Original Research

The availability of perinatal care in rural communities produces better pregnancy outcomes than do perinatal systems that require rural women to seek maternity care in distant urban areas.^1-3 Unfortunately, rural maternity care has been affected by the loss of physicians who offer these services and by the closing of many rural hospitals’ maternity care units. Maintaining 24-hour operative obstetric capabilities is difficult in rural areas because they have an insufficient population base to support a physician trained in operative obstetrics. Another barrier is the lack of anesthesia services and operating room personnel.

The Guidelines for Perinatal Care developed by the American College of Obstetricians and Gynecologists (ACOG) and the American Academy of Pediatrics (AAP) state, “All hospitals offering labor and delivery services should be equipped to perform emergency cesarean delivery.” ⁴ Nevertheless, not all rural obstetric units can offer cesarean delivery and must transfer patients to a referral hospital for operative needs. Advisory panels in the United States and Canada have recommended similar models of rural perinatal care.^5-8 A Canadian panel estimated that 125 Canadian hospitals offer obstetric care without surgical facilities.

Studies of rural hospitals in Canada, Australia, and the United Kingdom that lack continuous on-site cesarean capability are limited by the small number of deliveries.^9-12 Most such studies are hospital based rather than population based and lack data on women who are transferred to outlying hospitals. The only population-based study that we identified found no evidence of adverse events caused by the lack of cesarean facilities; the sample size, however, was limited to 286 births.⁹

We studied all pregnancies occurring in a predominantly Native American region of New Mexico over a 5-year period to ascertain the safety of rural perinatal care based in a hospital without cesarean capability. Population-based and hospital-based outcomes are presented. This is the first study from a US community using this model of care.

Methods

We conducted an outcomes study using a historical cohort study design of all pregnancies beyond 20 weeks of gestation in the Zuni Pueblo and Ramah Navajo communities of northwestern New Mexico from 1992 to 1996. The perinatal services based at the Zuni-Ramah Indian Health Service (IHS) Hospital are the focus of this study. This 37-bed community hospital, staffed by family physicians and a part-time nurse-midwife, is part of an integrated perinatal system. The birthing unit has access to obstetrician-gynecologist (OBG) consultants at the Gallup Indian Medical Center (GIMC), 33 miles to the north, and perinatology and neonatology care in Albuquerque, 147 miles to the east. GIMC, the primary referral hospital and closest surgical facility, has an obstetric unit staffed by OBGs, family physicians, and nurse-mid-wives. Transportation time is 40 minutes by ground ambulance to GIMC or by fixed wing aircraft to Albuquerque.

The Zuni-Ramah Hospital limits intrapartum care to women designated as at low or moderate risk by criteria established by Zuni-Ramah family physicians and reviewed by GIMC OBGs. Criteria mandating transfer included prior cesarean, malpresentation, multiple gestation, intrauterine growth restriction, severe preeclampsia, placenta previa, significant vaginal bleeding, major fetal anomalies, anticipated preterm delivery (< 36 weeks), nonreassuring fetal heart tones (NRFHTs), and need for labor induction or augmentation with oxytocin. Women with gestational or type 2 diabetes who were well controlled could give birth at Zuni-Ramah unless they had end-organ damage or the fetus had known macrosomia. Physicians successfully completed the Advanced Life Support in Obstetrics (ALSO, ®American Academy of Family Physicians, 4th ed., 2000) course, attended weekly high-risk obstetric rounds, and performed quarterly reviews of obstetric complications. The family physicians performed vacuum-assisted deliveries, utilized amnioinfusion, and used continuous or intermittent fetal monitoring.

A review of the delivery and transfer records of the Zuni-Ramah Hospital and GIMC obstetric services revealed that there had been 1132 births of 1137 infants during the study period. The authors used a data collection form to review prenatal and newborn records from every birth. We reviewed intrapartum records for all births at the Zuni-Ramah and GIMC hospitals. We obtained discharge summaries from tertiary-care sites. We interviewed perinatal coordinators, public health nurses, and pediatric care providers to obtain information about patients who had received perinatal care outside of the IHS system.

The outcomes measured included perinatal mortality, neonatal morbidity, obstetric emergencies, intrapartum and antepartum transfers, and cesarean delivery rate. All obstetric emergencies originating at Zuni-Ramah Hospital were reviewed to determine whether the lack of surgical facilities had resulted in adverse outcomes. The physician’s notes were used to differentiate a NRFHT pattern requiring observation at a hospital with operative facilities from a truly worrisome pattern that required urgent intervention for fetal distress.

Births were defined as deliveries of infants at 20 weeks or more of estimated gestational age. We analyzed each birth in a multiple gestation individually. The population-based perinatal mortality rate was calculated from 20 weeks’ estimated gestational age to the 28th neonatal day. The Zuni-Ramah Hospital perinatal mortality rate was calculated by inclusion of all women delivered at Zuni-Ramah Hospital or transferred during labor. Approval for the study was obtained from the IHS Institutional Review Board and the Zuni Tribal Council.

Results

Study population

We identified 1137 births to 945 women between 1992 and 1996. Zuni and Navajo births were 66.9% and 30.8%, respectively; 30% of women were primiparous and 70%, multiparous. We found that 10.4% of births had occurred in women older than 35 years and 7.8% in women younger than 18 years. Regarding prenatal care, 3.9% of women had received none; 43.0% had established prenatal care in the first trimester; 40.4%, in the second trimester; and 12.8%, in the third trimester.

Delivery sites and maternal transfers

The majority of women (64.4%, n = 732) gave birth at the Zuni-Ramah Hospital (Figure) or at GIMC (29.6%, n = 337). A small number (2.2%, n = 25) gave birth at a private hospital with surgical facilities in Gallup. Albuquerque tertiary-care hospitals were the sites of 3.2% (n = 36) of deliveries. Primary indications for tertiary care were prematurity and fetal anomalies. Seven (0.6%) deliveries occurred at other sites, including home and ambulance.

The antepartum transfers (Table 1) were required primarily for pregnancy complications requiring labor induction. Preeclampsia, diabetes, nonreassuring antepartum testing, and post dates patients accounted for 56.8% of the 290 transfers. The 107 intrapartum transfers were made predominantly for labor induction or augmentation (64.5%, n = 69), a concerning fetal heart tracing (15.9%, n = 17), or fetal malpresentation diagnosed during labor (8.4%, n = 9).

FIGURE
PREGNANCIES AT ZUNI-RAMAH HOSPITAL

TABLE 1
ANTEPARTUM AND INTRAPARTUM TRANSFERS FROM ZUNI-RAMAH HOSPITAL

Obstetric interventions

The total cesarean delivery rate (7.3%) was approximately one third the nationwide rate of 20.7% in 1996. The primary cesarean delivery rate (number of cesareans in women without prior cesarean divided by the number of women who have never had a cesarean) of 5.3% compares with a nationwide primary rate of 14.6%. The cesarean rate was 22.1% for antepartum transfers and 17.8% for intrapartum transfers. Operative vaginal delivery occurred in 5.4% of births, well below the nationwide rate of 9.4%. The induction rate of 13.8% is lower than the nationwide rate of 16.9%. The oxytocin augmentation rate of 7.7% is well below the nationwide rate of 16.9% in 1996.¹³ Parenteral narcotics were available at Zuni-Ramah; however, 81.4% of women elected to receive no labor analgesia. Epidural anesthesia was not available at Zuni-Ramah Hospital.

Perinatal mortality

The perinatal mortality rate for the population was 11.4 per 1000 births (95% CI, 5.1-17.8 by Poisson distribution), comparable to the 1991 nationwide peri-natal mortality rate of 12.8/1000.¹⁴ Nine of the 13 neonatal deaths were caused by intrauterine fetal demise before labor (Table 2). The Zuni-Ramah Hospital–based perinatal mortality rate of 1.2/1000 was comparable with the 1.3/1000 perinatal mortality rate for women in the National Birth Center study even though Zuni-Ramah Hospital accepts higher-risk patients.¹⁵

TABLE 2
PERINATAL MORTALITY IN ZUNI-RAMAH POPULATION

Neonatal morbidity

Measures of neonatal morbidity are summarized in Table 3. The frequency of 5-minute Apgar scores below 7, low birthweight, and prematurity compares favorably with 1996 US rates.¹³ The rate of assisted ventilation (intubation or bag-mask) for the entire population (4.6%, n = 52) is greater than the 1996 nationwide rate (2.9%), although the difference is of questionable clinical significance, since international studies have demonstrated a range for assisted ventilation of 1% to 10% of hospital births.¹⁶ Neonatal Intensive Care Unit (NICU) transfer occurred in 27 (2.4%) of deliveries from non-tertiary-care sites. Thirteen (1.8%) babies born at Zuni-Ramah were transferred to Albuquerque for NICU care because of respiratory distress (n = 10) or neonatal anomalies (n = 3). The 3 cases of low Apgar scores at Zuni-Ramah were attributed to pneumothorax, respiratory distress syndrome of prematurity, and sepsis with meconium aspiration.

TABLE 3
NEONATAL MORBIDITY IN ZUNI-RAMAH POPULATION, BASED ON LIVE BIRTHS

Obstetric risk factors

The study population had a greater incidence of pregnancy-induced hypertension (14.5% vs 2.6% by 1996 ACOG criteria¹⁷), chronic hypertension (2.7% vs 0.7%¹⁵), and diabetes (9.2% vs 2.6%¹⁵) than the average US obstetric population. Gestational diabetes was diagnosed according to National Diabetes Data Group criteria:¹⁸ 7.1% had gestational diabetes (class A₁ and A₂ ) and 2.1% had type 2 antepartum diabetes (classes B and C).

Outcomes of obstetric emergencies at zuni-ramah hospital

We reviewed all cases of placental abruption, uterine inversion, umbilical cord prolapse, and fetal distress at Zuni-Ramah Hospital to identify potentially preventable adverse outcomes caused by lack of operative facilities (Table W1). Umbilical cord prolapse and uterine inversion each occurred once and were appropriately managed, with excellent outcomes. In 3 of the 4 cases of placental abruption, there were clearly no adverse outcomes caused by lack of on-site operative facilities, as patients were expectantly managed upon arrival to the referral hospital (cases 3 and 4) or presented to Zuni-Ramah Hospital as an intrauterine demise (case 5).

The fourth patient with placental abruption (case 6) presented at Zuni-Ramah with vaginal bleeding, severe variable decelerations, and a 10-point drop from baseline hematocrit. She was scheduled to labor at GIMC because of a history of prior cesarean but presented to the Zuni-Ramah emergency room with vaginal bleeding. She was transferred to GIMC for an anticipated cesarean delivery; however, on arrival the patient rapidly progressed and gave birth to an infant vaginally with Apgar scores of 3 and 9. Her infant had a neonatal seizure and magnetic resonance imaging evidence of sagittal sinus thrombosis. The infant had a normal neurologic evaluation, developmental assessment, and electroencephalogram at 15 months.

We reviewed 5 cases of urgent transfer for fetal distress. These were differentiated from the 8 intrapartum transfers for NRFHTs based on the severity of fetal heart monitor tracings. Four of the 5 women who had been transferred for fetal distress gave birth to healthy infants vaginally more than 2 hours after arrival at the referral institution. One patient who was urgently transferred for repetitive late decelerations is discussed below.

Cesarean deliveries for fetal distress at referral hospitals

We reviewed all cases of cesarean deliveries for fetal distress (n = 10) at referral institutions to determine whether outcomes for any of the patients could potentially have been improved by having their cesarean deliveries earlier if operative facilities had been available at Zuni-Ramah Hospital or by being transferred earlier (Table W1). Seven of the 10 patients were transferred for preeclampsia, NRFHTs, or failure to progress. All had their cesareans for fetal distress many hours after arrival at the referral institution.

Two cases of cesarean delivery for fetal distress after transfer because of abruption were previously described. A patient presented in early labor with repetitive late decelerations and was urgently transferred to GIMC, where she underwent an immediate cesarean delivery. Her infant had Apgar scores of 1 and 7, an unremarkable neonatal course, and a normal 15-month developmental screen.

Discussion

Our outcomes demonstrate that with the use of appropriate screening criteria, childbirth can safely occur in institutions that lack surgical suites. The population-based perinatal mortality rate was similar to the nationwide rate. A review of obstetric emergencies and low Apgar scores among the 839 women laboring at Zuni-Ramah Hospital failed to identify adverse outcomes that might have been prevented if the hospital had had operative facilities. Cesarean rates were approximately one third the nationwide rate even though Zuni-Ramah patients had a higher prevalence of such risk factors as diabetes and preeclampsia.

Although they represented a high-risk obstetric population, 65% of women were able to give birth at Zuni-Ramah Hospital through use of the perinatal screening criteria. The 35% rate of transfer was caused largely by the need for oxytocin augmentation or induction. Only 21.6% of the women who were transferred for dysfunctional labor or premature rupture of membranes ultimately had a cesarean delivery. Oxytocin has not been permitted at Zuni-Ramah Hospital because of the ACOG guideline permitting oxytocin use only if “a physician capable of performing a cesarean delivery is readily available.”¹⁹ There are no studies addressing the safety of labor induction or augmentation without on-site cesarean capability.

Canadian guidelines for rural maternity care do not prohibit the use of prostaglandins or oxytocin at hospitals without operative facilities. A Consensus Conference on Obstetric Services in Rural or Remote Communities addressed the issue of labor induction or augmentation in hospitals without cesarean capability by stating, “If caring for a woman in labour is appropriate in the community, then caring for her during an augmented/induced labour is equally appropriate when there is support by trained local staff and resources.”²⁰ We concur that use of oxytocin in rural hospital units without operative facilities should be considered under well-defined clinical guidelines or research protocols.

Limitations

Our study’s limitations include lack of long-term neonatal outcomes, small size of the Zuni-Ramah population, an almost exclusively Native American population, and lack of examiner blinding during record review. Transfer rates may be increased in populations with higher rates of cesarean delivery or epidural anesthesia use. Alternatively, the high incidence of preeclampsia, chronic hypertension, and diabetes in these communities may have resulted in a higher proportion of induction. Umbilical cord pro-lapse and significant placental abruption are routinely treated by urgent cesarean delivery; therefore, obstetric literature on outcomes without immediate operative intervention is limited.^21,22 A larger study would be required to determine the potential increased neonatal morbidity or mortality resulting from delayed intervention.

Conclusions

The ACOG/AAP guideline requiring on-site surgical facilities and the ability to initiate a cesarean in 30 minutes is not based on evidence. Four small retrospective studies of emergency cesarean deliveries delayed for more than 30 minutes did not demonstrate adverse neonatal outcomes.^23-26 In our study population, no adverse outcomes (none in 839 births) were determined to have been caused by a lack of surgical facilities. Despite these excellent outcomes, the possibility always exists for outcomes that can be prevented by doing a rapid emergent cesarean delivery. Women deciding to give birth in facilities without operative capabilities should receive information regarding the risks and benefits of delivering there and should have access to other facilities. Provider discretion and patient choice must be respected to ensure community support of these birthing units. Practitioners at the rural units must have assurance that any patients who require an urgent transfer will be readily accepted.

Rural communities, medical providers, and health care facilities need to consider the overall effect of maintaining local maternity care units, as the loss of rural maternity care can increase the risk of adverse perinatal outcomes.^1-3 We concur with the Canadian panel that although maintenance of rural surgical and anesthesia capabilities is desirable, “good outcomes can be sustained within an integrated risk management system without local access to operative delivery.”⁸ Guidelines should be developed to permit rural hospitals without cesarean capability to provide maternity care as part of integrated perinatal systems with well-developed transport protocols and supportive referral institutions. Women living in rural areas should have the option to give birth near their homes in such units if they so desire.

Acknowledgments

The authors thank Robert Rhyne, MD, for editorial assistance in manuscript preparation and Betty Skipper, PhD, for statistical assistance. Current Zuni-Ramah Obstetric Guidelines are available at http://hsc.unm.edu/fcm/research/zuni.

The availability of perinatal care in rural communities produces better pregnancy outcomes than do perinatal systems that require rural women to seek maternity care in distant urban areas.^1-3 Unfortunately, rural maternity care has been affected by the loss of physicians who offer these services and by the closing of many rural hospitals’ maternity care units. Maintaining 24-hour operative obstetric capabilities is difficult in rural areas because they have an insufficient population base to support a physician trained in operative obstetrics. Another barrier is the lack of anesthesia services and operating room personnel.

The Guidelines for Perinatal Care developed by the American College of Obstetricians and Gynecologists (ACOG) and the American Academy of Pediatrics (AAP) state, “All hospitals offering labor and delivery services should be equipped to perform emergency cesarean delivery.” ⁴ Nevertheless, not all rural obstetric units can offer cesarean delivery and must transfer patients to a referral hospital for operative needs. Advisory panels in the United States and Canada have recommended similar models of rural perinatal care.^5-8 A Canadian panel estimated that 125 Canadian hospitals offer obstetric care without surgical facilities.

Studies of rural hospitals in Canada, Australia, and the United Kingdom that lack continuous on-site cesarean capability are limited by the small number of deliveries.^9-12 Most such studies are hospital based rather than population based and lack data on women who are transferred to outlying hospitals. The only population-based study that we identified found no evidence of adverse events caused by the lack of cesarean facilities; the sample size, however, was limited to 286 births.⁹

We studied all pregnancies occurring in a predominantly Native American region of New Mexico over a 5-year period to ascertain the safety of rural perinatal care based in a hospital without cesarean capability. Population-based and hospital-based outcomes are presented. This is the first study from a US community using this model of care.

Methods

We conducted an outcomes study using a historical cohort study design of all pregnancies beyond 20 weeks of gestation in the Zuni Pueblo and Ramah Navajo communities of northwestern New Mexico from 1992 to 1996. The perinatal services based at the Zuni-Ramah Indian Health Service (IHS) Hospital are the focus of this study. This 37-bed community hospital, staffed by family physicians and a part-time nurse-midwife, is part of an integrated perinatal system. The birthing unit has access to obstetrician-gynecologist (OBG) consultants at the Gallup Indian Medical Center (GIMC), 33 miles to the north, and perinatology and neonatology care in Albuquerque, 147 miles to the east. GIMC, the primary referral hospital and closest surgical facility, has an obstetric unit staffed by OBGs, family physicians, and nurse-mid-wives. Transportation time is 40 minutes by ground ambulance to GIMC or by fixed wing aircraft to Albuquerque.

The Zuni-Ramah Hospital limits intrapartum care to women designated as at low or moderate risk by criteria established by Zuni-Ramah family physicians and reviewed by GIMC OBGs. Criteria mandating transfer included prior cesarean, malpresentation, multiple gestation, intrauterine growth restriction, severe preeclampsia, placenta previa, significant vaginal bleeding, major fetal anomalies, anticipated preterm delivery (< 36 weeks), nonreassuring fetal heart tones (NRFHTs), and need for labor induction or augmentation with oxytocin. Women with gestational or type 2 diabetes who were well controlled could give birth at Zuni-Ramah unless they had end-organ damage or the fetus had known macrosomia. Physicians successfully completed the Advanced Life Support in Obstetrics (ALSO, ®American Academy of Family Physicians, 4th ed., 2000) course, attended weekly high-risk obstetric rounds, and performed quarterly reviews of obstetric complications. The family physicians performed vacuum-assisted deliveries, utilized amnioinfusion, and used continuous or intermittent fetal monitoring.

A review of the delivery and transfer records of the Zuni-Ramah Hospital and GIMC obstetric services revealed that there had been 1132 births of 1137 infants during the study period. The authors used a data collection form to review prenatal and newborn records from every birth. We reviewed intrapartum records for all births at the Zuni-Ramah and GIMC hospitals. We obtained discharge summaries from tertiary-care sites. We interviewed perinatal coordinators, public health nurses, and pediatric care providers to obtain information about patients who had received perinatal care outside of the IHS system.

The outcomes measured included perinatal mortality, neonatal morbidity, obstetric emergencies, intrapartum and antepartum transfers, and cesarean delivery rate. All obstetric emergencies originating at Zuni-Ramah Hospital were reviewed to determine whether the lack of surgical facilities had resulted in adverse outcomes. The physician’s notes were used to differentiate a NRFHT pattern requiring observation at a hospital with operative facilities from a truly worrisome pattern that required urgent intervention for fetal distress.

Births were defined as deliveries of infants at 20 weeks or more of estimated gestational age. We analyzed each birth in a multiple gestation individually. The population-based perinatal mortality rate was calculated from 20 weeks’ estimated gestational age to the 28th neonatal day. The Zuni-Ramah Hospital perinatal mortality rate was calculated by inclusion of all women delivered at Zuni-Ramah Hospital or transferred during labor. Approval for the study was obtained from the IHS Institutional Review Board and the Zuni Tribal Council.

Results

Study population

We identified 1137 births to 945 women between 1992 and 1996. Zuni and Navajo births were 66.9% and 30.8%, respectively; 30% of women were primiparous and 70%, multiparous. We found that 10.4% of births had occurred in women older than 35 years and 7.8% in women younger than 18 years. Regarding prenatal care, 3.9% of women had received none; 43.0% had established prenatal care in the first trimester; 40.4%, in the second trimester; and 12.8%, in the third trimester.

Delivery sites and maternal transfers

The majority of women (64.4%, n = 732) gave birth at the Zuni-Ramah Hospital (Figure) or at GIMC (29.6%, n = 337). A small number (2.2%, n = 25) gave birth at a private hospital with surgical facilities in Gallup. Albuquerque tertiary-care hospitals were the sites of 3.2% (n = 36) of deliveries. Primary indications for tertiary care were prematurity and fetal anomalies. Seven (0.6%) deliveries occurred at other sites, including home and ambulance.

The antepartum transfers (Table 1) were required primarily for pregnancy complications requiring labor induction. Preeclampsia, diabetes, nonreassuring antepartum testing, and post dates patients accounted for 56.8% of the 290 transfers. The 107 intrapartum transfers were made predominantly for labor induction or augmentation (64.5%, n = 69), a concerning fetal heart tracing (15.9%, n = 17), or fetal malpresentation diagnosed during labor (8.4%, n = 9).

FIGURE
PREGNANCIES AT ZUNI-RAMAH HOSPITAL

TABLE 1
ANTEPARTUM AND INTRAPARTUM TRANSFERS FROM ZUNI-RAMAH HOSPITAL

Obstetric interventions

The total cesarean delivery rate (7.3%) was approximately one third the nationwide rate of 20.7% in 1996. The primary cesarean delivery rate (number of cesareans in women without prior cesarean divided by the number of women who have never had a cesarean) of 5.3% compares with a nationwide primary rate of 14.6%. The cesarean rate was 22.1% for antepartum transfers and 17.8% for intrapartum transfers. Operative vaginal delivery occurred in 5.4% of births, well below the nationwide rate of 9.4%. The induction rate of 13.8% is lower than the nationwide rate of 16.9%. The oxytocin augmentation rate of 7.7% is well below the nationwide rate of 16.9% in 1996.¹³ Parenteral narcotics were available at Zuni-Ramah; however, 81.4% of women elected to receive no labor analgesia. Epidural anesthesia was not available at Zuni-Ramah Hospital.

Perinatal mortality

The perinatal mortality rate for the population was 11.4 per 1000 births (95% CI, 5.1-17.8 by Poisson distribution), comparable to the 1991 nationwide peri-natal mortality rate of 12.8/1000.¹⁴ Nine of the 13 neonatal deaths were caused by intrauterine fetal demise before labor (Table 2). The Zuni-Ramah Hospital–based perinatal mortality rate of 1.2/1000 was comparable with the 1.3/1000 perinatal mortality rate for women in the National Birth Center study even though Zuni-Ramah Hospital accepts higher-risk patients.¹⁵

TABLE 2
PERINATAL MORTALITY IN ZUNI-RAMAH POPULATION

Neonatal morbidity

Measures of neonatal morbidity are summarized in Table 3. The frequency of 5-minute Apgar scores below 7, low birthweight, and prematurity compares favorably with 1996 US rates.¹³ The rate of assisted ventilation (intubation or bag-mask) for the entire population (4.6%, n = 52) is greater than the 1996 nationwide rate (2.9%), although the difference is of questionable clinical significance, since international studies have demonstrated a range for assisted ventilation of 1% to 10% of hospital births.¹⁶ Neonatal Intensive Care Unit (NICU) transfer occurred in 27 (2.4%) of deliveries from non-tertiary-care sites. Thirteen (1.8%) babies born at Zuni-Ramah were transferred to Albuquerque for NICU care because of respiratory distress (n = 10) or neonatal anomalies (n = 3). The 3 cases of low Apgar scores at Zuni-Ramah were attributed to pneumothorax, respiratory distress syndrome of prematurity, and sepsis with meconium aspiration.

TABLE 3
NEONATAL MORBIDITY IN ZUNI-RAMAH POPULATION, BASED ON LIVE BIRTHS

Obstetric risk factors

The study population had a greater incidence of pregnancy-induced hypertension (14.5% vs 2.6% by 1996 ACOG criteria¹⁷), chronic hypertension (2.7% vs 0.7%¹⁵), and diabetes (9.2% vs 2.6%¹⁵) than the average US obstetric population. Gestational diabetes was diagnosed according to National Diabetes Data Group criteria:¹⁸ 7.1% had gestational diabetes (class A₁ and A₂ ) and 2.1% had type 2 antepartum diabetes (classes B and C).

Outcomes of obstetric emergencies at zuni-ramah hospital

We reviewed all cases of placental abruption, uterine inversion, umbilical cord prolapse, and fetal distress at Zuni-Ramah Hospital to identify potentially preventable adverse outcomes caused by lack of operative facilities (Table W1). Umbilical cord prolapse and uterine inversion each occurred once and were appropriately managed, with excellent outcomes. In 3 of the 4 cases of placental abruption, there were clearly no adverse outcomes caused by lack of on-site operative facilities, as patients were expectantly managed upon arrival to the referral hospital (cases 3 and 4) or presented to Zuni-Ramah Hospital as an intrauterine demise (case 5).

The fourth patient with placental abruption (case 6) presented at Zuni-Ramah with vaginal bleeding, severe variable decelerations, and a 10-point drop from baseline hematocrit. She was scheduled to labor at GIMC because of a history of prior cesarean but presented to the Zuni-Ramah emergency room with vaginal bleeding. She was transferred to GIMC for an anticipated cesarean delivery; however, on arrival the patient rapidly progressed and gave birth to an infant vaginally with Apgar scores of 3 and 9. Her infant had a neonatal seizure and magnetic resonance imaging evidence of sagittal sinus thrombosis. The infant had a normal neurologic evaluation, developmental assessment, and electroencephalogram at 15 months.

We reviewed 5 cases of urgent transfer for fetal distress. These were differentiated from the 8 intrapartum transfers for NRFHTs based on the severity of fetal heart monitor tracings. Four of the 5 women who had been transferred for fetal distress gave birth to healthy infants vaginally more than 2 hours after arrival at the referral institution. One patient who was urgently transferred for repetitive late decelerations is discussed below.

Cesarean deliveries for fetal distress at referral hospitals

We reviewed all cases of cesarean deliveries for fetal distress (n = 10) at referral institutions to determine whether outcomes for any of the patients could potentially have been improved by having their cesarean deliveries earlier if operative facilities had been available at Zuni-Ramah Hospital or by being transferred earlier (Table W1). Seven of the 10 patients were transferred for preeclampsia, NRFHTs, or failure to progress. All had their cesareans for fetal distress many hours after arrival at the referral institution.

Two cases of cesarean delivery for fetal distress after transfer because of abruption were previously described. A patient presented in early labor with repetitive late decelerations and was urgently transferred to GIMC, where she underwent an immediate cesarean delivery. Her infant had Apgar scores of 1 and 7, an unremarkable neonatal course, and a normal 15-month developmental screen.

Discussion

Our outcomes demonstrate that with the use of appropriate screening criteria, childbirth can safely occur in institutions that lack surgical suites. The population-based perinatal mortality rate was similar to the nationwide rate. A review of obstetric emergencies and low Apgar scores among the 839 women laboring at Zuni-Ramah Hospital failed to identify adverse outcomes that might have been prevented if the hospital had had operative facilities. Cesarean rates were approximately one third the nationwide rate even though Zuni-Ramah patients had a higher prevalence of such risk factors as diabetes and preeclampsia.

Although they represented a high-risk obstetric population, 65% of women were able to give birth at Zuni-Ramah Hospital through use of the perinatal screening criteria. The 35% rate of transfer was caused largely by the need for oxytocin augmentation or induction. Only 21.6% of the women who were transferred for dysfunctional labor or premature rupture of membranes ultimately had a cesarean delivery. Oxytocin has not been permitted at Zuni-Ramah Hospital because of the ACOG guideline permitting oxytocin use only if “a physician capable of performing a cesarean delivery is readily available.”¹⁹ There are no studies addressing the safety of labor induction or augmentation without on-site cesarean capability.

Canadian guidelines for rural maternity care do not prohibit the use of prostaglandins or oxytocin at hospitals without operative facilities. A Consensus Conference on Obstetric Services in Rural or Remote Communities addressed the issue of labor induction or augmentation in hospitals without cesarean capability by stating, “If caring for a woman in labour is appropriate in the community, then caring for her during an augmented/induced labour is equally appropriate when there is support by trained local staff and resources.”²⁰ We concur that use of oxytocin in rural hospital units without operative facilities should be considered under well-defined clinical guidelines or research protocols.

Limitations

Our study’s limitations include lack of long-term neonatal outcomes, small size of the Zuni-Ramah population, an almost exclusively Native American population, and lack of examiner blinding during record review. Transfer rates may be increased in populations with higher rates of cesarean delivery or epidural anesthesia use. Alternatively, the high incidence of preeclampsia, chronic hypertension, and diabetes in these communities may have resulted in a higher proportion of induction. Umbilical cord pro-lapse and significant placental abruption are routinely treated by urgent cesarean delivery; therefore, obstetric literature on outcomes without immediate operative intervention is limited.^21,22 A larger study would be required to determine the potential increased neonatal morbidity or mortality resulting from delayed intervention.

Conclusions

The ACOG/AAP guideline requiring on-site surgical facilities and the ability to initiate a cesarean in 30 minutes is not based on evidence. Four small retrospective studies of emergency cesarean deliveries delayed for more than 30 minutes did not demonstrate adverse neonatal outcomes.^23-26 In our study population, no adverse outcomes (none in 839 births) were determined to have been caused by a lack of surgical facilities. Despite these excellent outcomes, the possibility always exists for outcomes that can be prevented by doing a rapid emergent cesarean delivery. Women deciding to give birth in facilities without operative capabilities should receive information regarding the risks and benefits of delivering there and should have access to other facilities. Provider discretion and patient choice must be respected to ensure community support of these birthing units. Practitioners at the rural units must have assurance that any patients who require an urgent transfer will be readily accepted.

Rural communities, medical providers, and health care facilities need to consider the overall effect of maintaining local maternity care units, as the loss of rural maternity care can increase the risk of adverse perinatal outcomes.^1-3 We concur with the Canadian panel that although maintenance of rural surgical and anesthesia capabilities is desirable, “good outcomes can be sustained within an integrated risk management system without local access to operative delivery.”⁸ Guidelines should be developed to permit rural hospitals without cesarean capability to provide maternity care as part of integrated perinatal systems with well-developed transport protocols and supportive referral institutions. Women living in rural areas should have the option to give birth near their homes in such units if they so desire.

Acknowledgments

The authors thank Robert Rhyne, MD, for editorial assistance in manuscript preparation and Betty Skipper, PhD, for statistical assistance. Current Zuni-Ramah Obstetric Guidelines are available at http://hsc.unm.edu/fcm/research/zuni.

The availability of perinatal care in rural communities produces better pregnancy outcomes than do perinatal systems that require rural women to seek maternity care in distant urban areas.^1-3 Unfortunately, rural maternity care has been affected by the loss of physicians who offer these services and by the closing of many rural hospitals’ maternity care units. Maintaining 24-hour operative obstetric capabilities is difficult in rural areas because they have an insufficient population base to support a physician trained in operative obstetrics. Another barrier is the lack of anesthesia services and operating room personnel.

The Guidelines for Perinatal Care developed by the American College of Obstetricians and Gynecologists (ACOG) and the American Academy of Pediatrics (AAP) state, “All hospitals offering labor and delivery services should be equipped to perform emergency cesarean delivery.” ⁴ Nevertheless, not all rural obstetric units can offer cesarean delivery and must transfer patients to a referral hospital for operative needs. Advisory panels in the United States and Canada have recommended similar models of rural perinatal care.^5-8 A Canadian panel estimated that 125 Canadian hospitals offer obstetric care without surgical facilities.

Studies of rural hospitals in Canada, Australia, and the United Kingdom that lack continuous on-site cesarean capability are limited by the small number of deliveries.^9-12 Most such studies are hospital based rather than population based and lack data on women who are transferred to outlying hospitals. The only population-based study that we identified found no evidence of adverse events caused by the lack of cesarean facilities; the sample size, however, was limited to 286 births.⁹

We studied all pregnancies occurring in a predominantly Native American region of New Mexico over a 5-year period to ascertain the safety of rural perinatal care based in a hospital without cesarean capability. Population-based and hospital-based outcomes are presented. This is the first study from a US community using this model of care.

Methods

We conducted an outcomes study using a historical cohort study design of all pregnancies beyond 20 weeks of gestation in the Zuni Pueblo and Ramah Navajo communities of northwestern New Mexico from 1992 to 1996. The perinatal services based at the Zuni-Ramah Indian Health Service (IHS) Hospital are the focus of this study. This 37-bed community hospital, staffed by family physicians and a part-time nurse-midwife, is part of an integrated perinatal system. The birthing unit has access to obstetrician-gynecologist (OBG) consultants at the Gallup Indian Medical Center (GIMC), 33 miles to the north, and perinatology and neonatology care in Albuquerque, 147 miles to the east. GIMC, the primary referral hospital and closest surgical facility, has an obstetric unit staffed by OBGs, family physicians, and nurse-mid-wives. Transportation time is 40 minutes by ground ambulance to GIMC or by fixed wing aircraft to Albuquerque.

The Zuni-Ramah Hospital limits intrapartum care to women designated as at low or moderate risk by criteria established by Zuni-Ramah family physicians and reviewed by GIMC OBGs. Criteria mandating transfer included prior cesarean, malpresentation, multiple gestation, intrauterine growth restriction, severe preeclampsia, placenta previa, significant vaginal bleeding, major fetal anomalies, anticipated preterm delivery (< 36 weeks), nonreassuring fetal heart tones (NRFHTs), and need for labor induction or augmentation with oxytocin. Women with gestational or type 2 diabetes who were well controlled could give birth at Zuni-Ramah unless they had end-organ damage or the fetus had known macrosomia. Physicians successfully completed the Advanced Life Support in Obstetrics (ALSO, ®American Academy of Family Physicians, 4th ed., 2000) course, attended weekly high-risk obstetric rounds, and performed quarterly reviews of obstetric complications. The family physicians performed vacuum-assisted deliveries, utilized amnioinfusion, and used continuous or intermittent fetal monitoring.

A review of the delivery and transfer records of the Zuni-Ramah Hospital and GIMC obstetric services revealed that there had been 1132 births of 1137 infants during the study period. The authors used a data collection form to review prenatal and newborn records from every birth. We reviewed intrapartum records for all births at the Zuni-Ramah and GIMC hospitals. We obtained discharge summaries from tertiary-care sites. We interviewed perinatal coordinators, public health nurses, and pediatric care providers to obtain information about patients who had received perinatal care outside of the IHS system.

The outcomes measured included perinatal mortality, neonatal morbidity, obstetric emergencies, intrapartum and antepartum transfers, and cesarean delivery rate. All obstetric emergencies originating at Zuni-Ramah Hospital were reviewed to determine whether the lack of surgical facilities had resulted in adverse outcomes. The physician’s notes were used to differentiate a NRFHT pattern requiring observation at a hospital with operative facilities from a truly worrisome pattern that required urgent intervention for fetal distress.

Births were defined as deliveries of infants at 20 weeks or more of estimated gestational age. We analyzed each birth in a multiple gestation individually. The population-based perinatal mortality rate was calculated from 20 weeks’ estimated gestational age to the 28th neonatal day. The Zuni-Ramah Hospital perinatal mortality rate was calculated by inclusion of all women delivered at Zuni-Ramah Hospital or transferred during labor. Approval for the study was obtained from the IHS Institutional Review Board and the Zuni Tribal Council.

Results

Study population

We identified 1137 births to 945 women between 1992 and 1996. Zuni and Navajo births were 66.9% and 30.8%, respectively; 30% of women were primiparous and 70%, multiparous. We found that 10.4% of births had occurred in women older than 35 years and 7.8% in women younger than 18 years. Regarding prenatal care, 3.9% of women had received none; 43.0% had established prenatal care in the first trimester; 40.4%, in the second trimester; and 12.8%, in the third trimester.

Delivery sites and maternal transfers

The majority of women (64.4%, n = 732) gave birth at the Zuni-Ramah Hospital (Figure) or at GIMC (29.6%, n = 337). A small number (2.2%, n = 25) gave birth at a private hospital with surgical facilities in Gallup. Albuquerque tertiary-care hospitals were the sites of 3.2% (n = 36) of deliveries. Primary indications for tertiary care were prematurity and fetal anomalies. Seven (0.6%) deliveries occurred at other sites, including home and ambulance.

The antepartum transfers (Table 1) were required primarily for pregnancy complications requiring labor induction. Preeclampsia, diabetes, nonreassuring antepartum testing, and post dates patients accounted for 56.8% of the 290 transfers. The 107 intrapartum transfers were made predominantly for labor induction or augmentation (64.5%, n = 69), a concerning fetal heart tracing (15.9%, n = 17), or fetal malpresentation diagnosed during labor (8.4%, n = 9).

FIGURE
PREGNANCIES AT ZUNI-RAMAH HOSPITAL

TABLE 1
ANTEPARTUM AND INTRAPARTUM TRANSFERS FROM ZUNI-RAMAH HOSPITAL

Obstetric interventions

The total cesarean delivery rate (7.3%) was approximately one third the nationwide rate of 20.7% in 1996. The primary cesarean delivery rate (number of cesareans in women without prior cesarean divided by the number of women who have never had a cesarean) of 5.3% compares with a nationwide primary rate of 14.6%. The cesarean rate was 22.1% for antepartum transfers and 17.8% for intrapartum transfers. Operative vaginal delivery occurred in 5.4% of births, well below the nationwide rate of 9.4%. The induction rate of 13.8% is lower than the nationwide rate of 16.9%. The oxytocin augmentation rate of 7.7% is well below the nationwide rate of 16.9% in 1996.¹³ Parenteral narcotics were available at Zuni-Ramah; however, 81.4% of women elected to receive no labor analgesia. Epidural anesthesia was not available at Zuni-Ramah Hospital.

Perinatal mortality

The perinatal mortality rate for the population was 11.4 per 1000 births (95% CI, 5.1-17.8 by Poisson distribution), comparable to the 1991 nationwide peri-natal mortality rate of 12.8/1000.¹⁴ Nine of the 13 neonatal deaths were caused by intrauterine fetal demise before labor (Table 2). The Zuni-Ramah Hospital–based perinatal mortality rate of 1.2/1000 was comparable with the 1.3/1000 perinatal mortality rate for women in the National Birth Center study even though Zuni-Ramah Hospital accepts higher-risk patients.¹⁵

TABLE 2
PERINATAL MORTALITY IN ZUNI-RAMAH POPULATION

Neonatal morbidity

Measures of neonatal morbidity are summarized in Table 3. The frequency of 5-minute Apgar scores below 7, low birthweight, and prematurity compares favorably with 1996 US rates.¹³ The rate of assisted ventilation (intubation or bag-mask) for the entire population (4.6%, n = 52) is greater than the 1996 nationwide rate (2.9%), although the difference is of questionable clinical significance, since international studies have demonstrated a range for assisted ventilation of 1% to 10% of hospital births.¹⁶ Neonatal Intensive Care Unit (NICU) transfer occurred in 27 (2.4%) of deliveries from non-tertiary-care sites. Thirteen (1.8%) babies born at Zuni-Ramah were transferred to Albuquerque for NICU care because of respiratory distress (n = 10) or neonatal anomalies (n = 3). The 3 cases of low Apgar scores at Zuni-Ramah were attributed to pneumothorax, respiratory distress syndrome of prematurity, and sepsis with meconium aspiration.

TABLE 3
NEONATAL MORBIDITY IN ZUNI-RAMAH POPULATION, BASED ON LIVE BIRTHS

Obstetric risk factors

The study population had a greater incidence of pregnancy-induced hypertension (14.5% vs 2.6% by 1996 ACOG criteria¹⁷), chronic hypertension (2.7% vs 0.7%¹⁵), and diabetes (9.2% vs 2.6%¹⁵) than the average US obstetric population. Gestational diabetes was diagnosed according to National Diabetes Data Group criteria:¹⁸ 7.1% had gestational diabetes (class A₁ and A₂ ) and 2.1% had type 2 antepartum diabetes (classes B and C).

Outcomes of obstetric emergencies at zuni-ramah hospital

We reviewed all cases of placental abruption, uterine inversion, umbilical cord prolapse, and fetal distress at Zuni-Ramah Hospital to identify potentially preventable adverse outcomes caused by lack of operative facilities (Table W1). Umbilical cord prolapse and uterine inversion each occurred once and were appropriately managed, with excellent outcomes. In 3 of the 4 cases of placental abruption, there were clearly no adverse outcomes caused by lack of on-site operative facilities, as patients were expectantly managed upon arrival to the referral hospital (cases 3 and 4) or presented to Zuni-Ramah Hospital as an intrauterine demise (case 5).

The fourth patient with placental abruption (case 6) presented at Zuni-Ramah with vaginal bleeding, severe variable decelerations, and a 10-point drop from baseline hematocrit. She was scheduled to labor at GIMC because of a history of prior cesarean but presented to the Zuni-Ramah emergency room with vaginal bleeding. She was transferred to GIMC for an anticipated cesarean delivery; however, on arrival the patient rapidly progressed and gave birth to an infant vaginally with Apgar scores of 3 and 9. Her infant had a neonatal seizure and magnetic resonance imaging evidence of sagittal sinus thrombosis. The infant had a normal neurologic evaluation, developmental assessment, and electroencephalogram at 15 months.

We reviewed 5 cases of urgent transfer for fetal distress. These were differentiated from the 8 intrapartum transfers for NRFHTs based on the severity of fetal heart monitor tracings. Four of the 5 women who had been transferred for fetal distress gave birth to healthy infants vaginally more than 2 hours after arrival at the referral institution. One patient who was urgently transferred for repetitive late decelerations is discussed below.

Cesarean deliveries for fetal distress at referral hospitals

We reviewed all cases of cesarean deliveries for fetal distress (n = 10) at referral institutions to determine whether outcomes for any of the patients could potentially have been improved by having their cesarean deliveries earlier if operative facilities had been available at Zuni-Ramah Hospital or by being transferred earlier (Table W1). Seven of the 10 patients were transferred for preeclampsia, NRFHTs, or failure to progress. All had their cesareans for fetal distress many hours after arrival at the referral institution.

Two cases of cesarean delivery for fetal distress after transfer because of abruption were previously described. A patient presented in early labor with repetitive late decelerations and was urgently transferred to GIMC, where she underwent an immediate cesarean delivery. Her infant had Apgar scores of 1 and 7, an unremarkable neonatal course, and a normal 15-month developmental screen.

Discussion

Our outcomes demonstrate that with the use of appropriate screening criteria, childbirth can safely occur in institutions that lack surgical suites. The population-based perinatal mortality rate was similar to the nationwide rate. A review of obstetric emergencies and low Apgar scores among the 839 women laboring at Zuni-Ramah Hospital failed to identify adverse outcomes that might have been prevented if the hospital had had operative facilities. Cesarean rates were approximately one third the nationwide rate even though Zuni-Ramah patients had a higher prevalence of such risk factors as diabetes and preeclampsia.

Although they represented a high-risk obstetric population, 65% of women were able to give birth at Zuni-Ramah Hospital through use of the perinatal screening criteria. The 35% rate of transfer was caused largely by the need for oxytocin augmentation or induction. Only 21.6% of the women who were transferred for dysfunctional labor or premature rupture of membranes ultimately had a cesarean delivery. Oxytocin has not been permitted at Zuni-Ramah Hospital because of the ACOG guideline permitting oxytocin use only if “a physician capable of performing a cesarean delivery is readily available.”¹⁹ There are no studies addressing the safety of labor induction or augmentation without on-site cesarean capability.

Canadian guidelines for rural maternity care do not prohibit the use of prostaglandins or oxytocin at hospitals without operative facilities. A Consensus Conference on Obstetric Services in Rural or Remote Communities addressed the issue of labor induction or augmentation in hospitals without cesarean capability by stating, “If caring for a woman in labour is appropriate in the community, then caring for her during an augmented/induced labour is equally appropriate when there is support by trained local staff and resources.”²⁰ We concur that use of oxytocin in rural hospital units without operative facilities should be considered under well-defined clinical guidelines or research protocols.

Limitations

Our study’s limitations include lack of long-term neonatal outcomes, small size of the Zuni-Ramah population, an almost exclusively Native American population, and lack of examiner blinding during record review. Transfer rates may be increased in populations with higher rates of cesarean delivery or epidural anesthesia use. Alternatively, the high incidence of preeclampsia, chronic hypertension, and diabetes in these communities may have resulted in a higher proportion of induction. Umbilical cord pro-lapse and significant placental abruption are routinely treated by urgent cesarean delivery; therefore, obstetric literature on outcomes without immediate operative intervention is limited.^21,22 A larger study would be required to determine the potential increased neonatal morbidity or mortality resulting from delayed intervention.

Conclusions

The ACOG/AAP guideline requiring on-site surgical facilities and the ability to initiate a cesarean in 30 minutes is not based on evidence. Four small retrospective studies of emergency cesarean deliveries delayed for more than 30 minutes did not demonstrate adverse neonatal outcomes.^23-26 In our study population, no adverse outcomes (none in 839 births) were determined to have been caused by a lack of surgical facilities. Despite these excellent outcomes, the possibility always exists for outcomes that can be prevented by doing a rapid emergent cesarean delivery. Women deciding to give birth in facilities without operative capabilities should receive information regarding the risks and benefits of delivering there and should have access to other facilities. Provider discretion and patient choice must be respected to ensure community support of these birthing units. Practitioners at the rural units must have assurance that any patients who require an urgent transfer will be readily accepted.

Rural communities, medical providers, and health care facilities need to consider the overall effect of maintaining local maternity care units, as the loss of rural maternity care can increase the risk of adverse perinatal outcomes.^1-3 We concur with the Canadian panel that although maintenance of rural surgical and anesthesia capabilities is desirable, “good outcomes can be sustained within an integrated risk management system without local access to operative delivery.”⁸ Guidelines should be developed to permit rural hospitals without cesarean capability to provide maternity care as part of integrated perinatal systems with well-developed transport protocols and supportive referral institutions. Women living in rural areas should have the option to give birth near their homes in such units if they so desire.

Acknowledgments

The authors thank Robert Rhyne, MD, for editorial assistance in manuscript preparation and Betty Skipper, PhD, for statistical assistance. Current Zuni-Ramah Obstetric Guidelines are available at http://hsc.unm.edu/fcm/research/zuni.

Postherpetic neuralgia (PHN), the most common complication of herpes zoster, is much more prevalent among adults older than 50 years than in younger people.^1,2 The largest English-language prospective study of patients presenting with zoster suggests that the average family physician can expect to see 4 cases of zoster per year and 1 case of PHN lasting more than 3 months every 3 years.³ Among placebo cohorts from randomized controlled trials (RCTs) of acute zoster treatment, the incidence of pain at 3 months has been reported as 17% to 60%; at 6 months, 5% to 39%.⁴ There is limited evidence that therapies for acute zoster have an impact on PHN.⁴

This review addresses therapies to reduce pain or improve quality of life in patients with PHN. The condition has been variously defined in terms of timing (either following resolution of acute zoster [rash healing] or a defined time after onset of zoster), duration (any time after zoster or a minimum of 6 months after zoster), and type of pain (such as lancinating pain or allodynia [pain caused by a stimulus that does not nor mally provoke pain]).^5,6 PHN may include a spectrum of presentations, from brief intermittent mild pain that resolves spontaneously to chronic persistent disabling pain recalcitrant to multiple therapies. To avoid missing potentially relevant findings, we defined PHN broadly as any pain after cutaneous healing of zoster.

Methods

Search strategy

Medline (1966 to present) was searched on October 19, 2000. The search combined the terms “post-herpetic or postherpetic” and “neuralgia or neuropathy or pain” and publication type “clinical trial (including phases I to IV) or controlled clinical trial or randomized controlled trial.” The Cochrane Controlled Trials Registry 2000, Issue 3, was searched with the same terms. Current Contents was searched to identify more recent references. We also identified trials through article reference lists (from included trials and 40 reviews), contact of authors and content experts, and the Food and Drug Administration (FDA) Web site.

Selection criteria

Inclusion criteria for this review were RCTs that enrolled patients with PHN (history of zoster, pain in the dermatomal distribution of the zoster rash, and pain persisting or occurring after resolution of the zoster rash), addressed relevant end points (pain resolution, pain severity, effect on quality of life), and had full reports available in English. Since responses to initial therapy may change over time, we included only trials with evaluation periods lasting more than 24 hours.

The authors independently evaluated trials meeting these inclusion criteria for quality and independently extracted data. Quality was evaluated using the Jadad scale,⁷ which addresses selected criteria (randomization technique, allocation concealment, blinding, accounting of dropouts) and rates methodologic quality on a 5-point scale, with 5 representing the highest score. Differences were resolved through discussion. Trials scoring only 1 point were excluded except for 2 instances noted in our discussion.

Results

Searching identified 186 potential trials, of which 92 were excluded as irrelevant on the basis of titles and abstracts alone. Of the 94 citations reviewed in greater detail, 64 were excluded for the following reasons: not describing a trial (10), not describing a trial of treatment for PHN (10), describing an uncontrolled trial (7), no randomization (7), evaluation period limited to 24 hours or less (13), duplicate publication (4), language other than English (7), not providing results specifically for patients with PHN (3), and available only in abstract form (3). One unpublished trial of mexiletine was identified through a review by Hanania and Brietstein; Dr Hanania informed us, however, that this trial had been stopped early because the treatment group had experienced serious side effects. One controlled trial was excluded because correspondence with the investigator did not confirm that it had been randomized. One trial was excluded because it included only 6 patients with PHN. (A list of excluded studies is available in Table W1.)

Of the remaining 27 trials reviewed for methodologic quality, most (16) received a Jadad score of 4. The 2 authors had substantial agreement on quality ratings ( = 0.75). Table W2 provides details of treatment regimens, quality ratings, ages of subjects, and duration of PHN. One trial with a Jadad score of 1 was excluded.⁸ Most subjects were elderly and had had PHN for longer than 6 months.

Topical therapies

Topical therapies evaluated were lidocaine, capsaicin, and benzydamine (Table 1). Lidocaine patch therapy is the only agent with a specific FDA indication for PHN. We found few trials supporting the FDA approval. The only published RCT of relatively unselected patients with PHN (n = 35) showed significant benefit versus placebo but was excluded because evaluation sessions had been limited to 12 hours.⁹ We reviewed a report of an unpublished RCT comparing lidocaine patch with vehicle placebo used in the application for FDA approval.¹⁰ This trial found a large, statistically significant reduction in pain scores with placebo throughout the 3- to 4-week trial. This trial found a similar statistically significant reduction in pain scores with lidocaine patch and no significant difference comparing lidocaine patch with placebo.

Three findings in the unpublished trial were used to support arguments for efficacy: (1) a statistically significant difference in the pain relief score at the final visit; (2) differences in allodynia (based on investigators’ sensory skin testing, described as stroking the maximally painful area with a foam brush and recording the pain scale rating) at the beginning of the trial; and (3) a greater increase in pain scores among lidocaine subjects upon trial conclusion (ie, after stopping study medication). The clinical relevance of these 3 findings is unclear.

The FDA declined to approve lidocaine patch therapy on the basis of these 2 studies and required an additional trial to demonstrate benefit. An “enriched enrollment study” involved subjects who had used lidocaine patch for at least 1 month and received at least moderate relief but had pain without the patch.¹¹ Lidocaine patch was clearly effective in this highly selected cohort.

Capsaicin 0.075% cream was effective in 2 trials of patients with severe refractory PHN.^12,13 The benefit appeared modest in the larger trial (pain was eliminated or nearly eliminated in less than 20% of capsaicin patients) and greater in the smaller trial. Blinding had limited efficacy because of the stinging effect of capsaicin.

Benzydamine cream, an antiprostaglandin, was not effective in a 2-week crossover trial.¹⁴ The cream showed a nonsignificant trend for pain reduction in an earlier 2-week crossover trial.¹⁵

TABLE 1

Oral therapies

Oral therapies evaluated were tricyclic antidepressants, gabapentin, oxycodone, tramadol, dextromethorphan, memantine, acyclovir, lorazepam, and fluphenazine (Table 1).

Tricyclic antidepressants have been shown to be effective in multiple small short-term crossover trials. Amitriptyline was highly effective in 2 placebo-controlled trials.^16,17 In 1 of these trials, amitriptyline was more effective than lorazepam.¹⁶ In another trial, amitriptyline was more effective than fluphenazine (a phenothiazine) and glycopyrrolate placebo.¹⁸ Nortriptyline was as effective as amitriptyline in a comparison trial,¹⁹ while maprotiline was not.²⁰ Desipramine was highly effective in a trial using benztropine as an “active placebo” in that the anticholinergic properties of benztropine were used to match the side effects of desipramine.²¹ In all these studies, the analgesic effects of tricyclic antidepressants appeared independent of antidepressant effects. No randomized trial data were collected to assess the use of antidepressants for longer than 8 weeks.

Gabapentin, an anticonvulsant, was effective in a single large placebo-controlled trial.²² The number needed to treat (NNT) was 3.2 for the outcome of moderate or better pain relief and 13.9 for the outcome of no pain during the eighth week of treatment. The proportion of patients whose pain was much improved or who had no pain was not reported.

Controlled-release oxycodone was effective in a crossover trial in which 45% of patients had previously used opioids.²³ Tramadol may be effective but was not compared against placebo.²⁴ High-dose dextromethorphan, an N-methyl-D-aspartate (NMDA) receptor antagonist, was not shown to be effective for PHN in a small crossover trial.²⁵ Memantine, another NMDA antagonist, was also ineffective.²⁶ Acyclovir did not show any greater efficacy than placebo in a small trial.²⁷ Lorazepam and fluphenazine did not show statistically significant benefit in comparison with placebo in the amitriptyline trials.^16,18

Other therapies

Other therapies evaluated were vincristine iontophoresis, acupuncture, intrathecal methylprednisolone, and subcutaneous administration of a mixture of gangliosides (Table 1).

Iontophoresis is a process whereby topical medications are applied via electricity. Vincristine iontophoresis was no more effective than saline iontophoresis in one small trial.²⁸ Vincristine and dimethylsulfoxide iontophoresis was effective at reducing but not eliminating pain in another small trial.²⁹ Dimethylsulfoxide may have an independent analgesic effect.³⁰

Acupuncture was no more effective than mock transcutaneous electrical nerve stimulation (TENS) in 1 trial,³¹ while a smaller trial suggested a short-term effect.³²

Intrathecal methylprednisolone acetate plus lidocaine was highly effective for achieving good or excellent results (pain relief > 50%) in patients with longstanding PHN refractory to multiple conventional therapies.³³ All patients whose response to methylprednisolone was poor (8%) had had PHN for more than 5 years. The intrathecal route appears more effective than the epidural route of administration.³⁴

A mixture of gangliosides given by subcutaneous injection was more effective than placebo, but poor tolerability and derivation from bovine brain tissue severely limit its acceptability.³⁵

Sympathetic blocks using bupivacaine were more effective than intravenous lidocaine infusions in 1 trial,³⁶ but results were not reported in a fashion that conveys the proportion of patients who improved significantly.

Discussion

Effective therapies

The therapy for which evidence for efficacy is best is tricyclic antidepressants. Three placebo-controlled RCTs demonstrated that only 2 to 3 patients with PHN need to be treated to achieve 1 good or excellent result (NNT = 2-3). Since none of these studies lasted longer than 8 weeks, the long-term efficacy of antidepressants for the treatment of PHN is unknown. Follow-up of 10 patients who did well in 1 antidepressant trial found that only 2 patients were still doing well at 2 years.¹⁹

Other therapies shown to be effective in 1 or 2 trials are topical capsaicin 0.075%, gabapentin, and controlled-release oxycodone. For these studies, it is difficult to determine the number needed to treat for “meaningful” clinical benefit, although gabapentin demonstrated superiority to placebo in numerous quality-of-life measures.

For patients with severe PHN refractory to other treatments, 2 trials support benefit from intrathecal methylprednisolone and 1 trial suggests benefit from bupivacaine sympathetic blocks. Cost data for selected therapies are presented in Table 2.

TABLE 2
COSTS OF SELECTED DRUG THERAPIES

Therapies not proved effective

Therapies of uncertain benefit that have not been adequately studied in randomized controlled trials include lidocaine patch, benzydamine cream, tramadol, and vincristine (and/or dimethylsulfoxide) iontophoresis.

Therapies unlikely to be beneficial based on single trials include lorazepam, fluphenazine, dextromethorphan, memantine, acyclovir, and acupuncture. Most of the negative trials did not report power; therefore, potential benefits of these treatments cannot be excluded.

Safety and tolerability

The rates of adverse effects are high in all effective oral and topical therapies (Table 1). This situation is of special concern in elderly patients who have comorbid conditions and are taking multiple medications. Two tricyclic antidepressant trials^16,21 report a decreased incidence of side effects over time. The researchers emphasize the importance of starting at the lowest available dose with oral therapies and titrating slowly as indicated and tolerated.

No clinical complications were observed in the intrathecal steroid trials; specific side effects were not reported.

Lidocaine patch therapy has been promoted as causing clinically insignificant serum levels, no systemic side effects, and no drug–drug interactions.¹¹ However, the largest lidocaine patch RCT was too small to rule out significant but uncommon risks such as ventricular arrhythmia.¹⁰ One death that could potentially be attributed to lidocaine absorption occurred in a patient with diffuse vascular disease who was on chronic hemodialysis for renal failure. Blood lidocaine levels were not obtained because venous access was poor.

Limitations

Variations in outcomes limit comparative conclusions. Sindrup and Jensen³⁷ reviewed treatments for neuropathic pain and presented data based on a successful outcome defined as 50% reduction in pain scores, 50% pain relief, or categorical ratings of excellent, good, or moderate pain relief. Most studies used visual analog scales, but it is not clear that a 50% reduction in these measures is equivalent to clinically meaningful benefits at all levels of pain. We attempted to determine NNT data based on clearly meaningful outcomes such as “excellent or good,” “complete,” or “marked” pain relief as distinct from “moderate” or “some,” but found these data unavailable in most reports. Quality-of-life measures, such as sleep and disability ratings, may be more important than measures of pain, but were reported in only one third of the trials.

We may have failed to include relevant trials. We identified 7 potentially relevant non–English language studies. Five had no abstracts available.^38-42 One single-blind trial reported reduced pain scores with topical prostaglandin E1 dissolved in Vaseline.⁴³ One trial compared iontophoresis with lidocaine and iontophoresis with 3 different calcium channel blockers in 10 patients. The authors found that all 4 treatments reduced pain, but had not included a placebo control group.⁴⁴ We also contacted 15 content experts to identify reports of unpublished trials; none of the 6 responses received identified such reports.

The evidence base for treatment of PHN is limited. Among the RCTs reviewed, 78% enrolled 50 or fewer patients. Because most of the subjects had PHN lasting longer than 1 year, our conclusions may not apply to patients with PHN of shorter duration. The latter group represents the majority of subjects with PHN presenting to primary care physicians.

We used the Jadad scale⁷ as an attempt to quantitatively assess the methodologic quality of the studies we reviewed. In general, explicit validity checklists with summary scores have not consistently been shown to provide more reliable assessments of validity than qualitative assessments.^45-4 The Jadad scale is the first validity checklist that has some rigorous evidence supporting its use,^7,48 although its inter-rater reliability has been questioned.⁴⁹ We found the Jadad scale had an inherent bias against therapies that could not be adequately double-blinded. Thus 2 trials with a score of 1 were included.^29,36 In 1 trial of vincristine iontophoresis, the authors described the trial as single-blinded and provided ample explanation of why double-blinding was not achieved.²⁹ In 1 trial of sympathetic blocks, the treatment studied included an invasive procedure; therefore, there was no apparent way for the procedure itself to be double-blinded.³⁶

The Jadad scale does not account for some threats to validity of included studies. We encountered numerous methodologic flaws, such as lack of intention to treat analysis in parallel trials (11) and lack of washout periods in crossover trials (4). Further limitations to interpretation of selected study results included potentially significant baseline differences between groups (8), small numbers, and short study durations. A list of the studies with these specific methodologic concerns is available in Table W3.

Conclusions

For patients with PHN lasting less than 6 months, spontaneous resolution is common and treatment decisions are largely empiric and not evidence based. For PHN of longer duration, treatments shown to be more effective than placebo include tricyclic antidepressants, topical capsaicin 0.075%, gabapentin, and controlled-release oxycodone. These treatments all have adverse effects or costs that need to be considered on an individual basis. Lidocaine patch therapy may be safer for most patients but may be no more effective than placebo and is not suitable for patients with trigeminal PHN. Patients with PHN refractory to the currently available and studied topical and oral agents should be considered for intrathecal steroid therapy.

Acknowledgments

The authors wish to thank Susan Meadows, MLS, Susan Elliott, MLS, Stacey Rautzhan, and Steve Calloway, RPh, for their assistance and Sigurdur Helgason, MD, Carin Reust, MD, Steven Zweig, MD, MSPH, and Alan Adelman, MD, MS, for editorial review.

Postherpetic neuralgia (PHN), the most common complication of herpes zoster, is much more prevalent among adults older than 50 years than in younger people.^1,2 The largest English-language prospective study of patients presenting with zoster suggests that the average family physician can expect to see 4 cases of zoster per year and 1 case of PHN lasting more than 3 months every 3 years.³ Among placebo cohorts from randomized controlled trials (RCTs) of acute zoster treatment, the incidence of pain at 3 months has been reported as 17% to 60%; at 6 months, 5% to 39%.⁴ There is limited evidence that therapies for acute zoster have an impact on PHN.⁴

This review addresses therapies to reduce pain or improve quality of life in patients with PHN. The condition has been variously defined in terms of timing (either following resolution of acute zoster [rash healing] or a defined time after onset of zoster), duration (any time after zoster or a minimum of 6 months after zoster), and type of pain (such as lancinating pain or allodynia [pain caused by a stimulus that does not nor mally provoke pain]).^5,6 PHN may include a spectrum of presentations, from brief intermittent mild pain that resolves spontaneously to chronic persistent disabling pain recalcitrant to multiple therapies. To avoid missing potentially relevant findings, we defined PHN broadly as any pain after cutaneous healing of zoster.

Methods

Search strategy

Medline (1966 to present) was searched on October 19, 2000. The search combined the terms “post-herpetic or postherpetic” and “neuralgia or neuropathy or pain” and publication type “clinical trial (including phases I to IV) or controlled clinical trial or randomized controlled trial.” The Cochrane Controlled Trials Registry 2000, Issue 3, was searched with the same terms. Current Contents was searched to identify more recent references. We also identified trials through article reference lists (from included trials and 40 reviews), contact of authors and content experts, and the Food and Drug Administration (FDA) Web site.

Selection criteria

Inclusion criteria for this review were RCTs that enrolled patients with PHN (history of zoster, pain in the dermatomal distribution of the zoster rash, and pain persisting or occurring after resolution of the zoster rash), addressed relevant end points (pain resolution, pain severity, effect on quality of life), and had full reports available in English. Since responses to initial therapy may change over time, we included only trials with evaluation periods lasting more than 24 hours.

The authors independently evaluated trials meeting these inclusion criteria for quality and independently extracted data. Quality was evaluated using the Jadad scale,⁷ which addresses selected criteria (randomization technique, allocation concealment, blinding, accounting of dropouts) and rates methodologic quality on a 5-point scale, with 5 representing the highest score. Differences were resolved through discussion. Trials scoring only 1 point were excluded except for 2 instances noted in our discussion.

Results

Searching identified 186 potential trials, of which 92 were excluded as irrelevant on the basis of titles and abstracts alone. Of the 94 citations reviewed in greater detail, 64 were excluded for the following reasons: not describing a trial (10), not describing a trial of treatment for PHN (10), describing an uncontrolled trial (7), no randomization (7), evaluation period limited to 24 hours or less (13), duplicate publication (4), language other than English (7), not providing results specifically for patients with PHN (3), and available only in abstract form (3). One unpublished trial of mexiletine was identified through a review by Hanania and Brietstein; Dr Hanania informed us, however, that this trial had been stopped early because the treatment group had experienced serious side effects. One controlled trial was excluded because correspondence with the investigator did not confirm that it had been randomized. One trial was excluded because it included only 6 patients with PHN. (A list of excluded studies is available in Table W1.)

Of the remaining 27 trials reviewed for methodologic quality, most (16) received a Jadad score of 4. The 2 authors had substantial agreement on quality ratings ( = 0.75). Table W2 provides details of treatment regimens, quality ratings, ages of subjects, and duration of PHN. One trial with a Jadad score of 1 was excluded.⁸ Most subjects were elderly and had had PHN for longer than 6 months.

Topical therapies

Topical therapies evaluated were lidocaine, capsaicin, and benzydamine (Table 1). Lidocaine patch therapy is the only agent with a specific FDA indication for PHN. We found few trials supporting the FDA approval. The only published RCT of relatively unselected patients with PHN (n = 35) showed significant benefit versus placebo but was excluded because evaluation sessions had been limited to 12 hours.⁹ We reviewed a report of an unpublished RCT comparing lidocaine patch with vehicle placebo used in the application for FDA approval.¹⁰ This trial found a large, statistically significant reduction in pain scores with placebo throughout the 3- to 4-week trial. This trial found a similar statistically significant reduction in pain scores with lidocaine patch and no significant difference comparing lidocaine patch with placebo.

Three findings in the unpublished trial were used to support arguments for efficacy: (1) a statistically significant difference in the pain relief score at the final visit; (2) differences in allodynia (based on investigators’ sensory skin testing, described as stroking the maximally painful area with a foam brush and recording the pain scale rating) at the beginning of the trial; and (3) a greater increase in pain scores among lidocaine subjects upon trial conclusion (ie, after stopping study medication). The clinical relevance of these 3 findings is unclear.

The FDA declined to approve lidocaine patch therapy on the basis of these 2 studies and required an additional trial to demonstrate benefit. An “enriched enrollment study” involved subjects who had used lidocaine patch for at least 1 month and received at least moderate relief but had pain without the patch.¹¹ Lidocaine patch was clearly effective in this highly selected cohort.

Capsaicin 0.075% cream was effective in 2 trials of patients with severe refractory PHN.^12,13 The benefit appeared modest in the larger trial (pain was eliminated or nearly eliminated in less than 20% of capsaicin patients) and greater in the smaller trial. Blinding had limited efficacy because of the stinging effect of capsaicin.

Benzydamine cream, an antiprostaglandin, was not effective in a 2-week crossover trial.¹⁴ The cream showed a nonsignificant trend for pain reduction in an earlier 2-week crossover trial.¹⁵

TABLE 1

Oral therapies

Oral therapies evaluated were tricyclic antidepressants, gabapentin, oxycodone, tramadol, dextromethorphan, memantine, acyclovir, lorazepam, and fluphenazine (Table 1).

Tricyclic antidepressants have been shown to be effective in multiple small short-term crossover trials. Amitriptyline was highly effective in 2 placebo-controlled trials.^16,17 In 1 of these trials, amitriptyline was more effective than lorazepam.¹⁶ In another trial, amitriptyline was more effective than fluphenazine (a phenothiazine) and glycopyrrolate placebo.¹⁸ Nortriptyline was as effective as amitriptyline in a comparison trial,¹⁹ while maprotiline was not.²⁰ Desipramine was highly effective in a trial using benztropine as an “active placebo” in that the anticholinergic properties of benztropine were used to match the side effects of desipramine.²¹ In all these studies, the analgesic effects of tricyclic antidepressants appeared independent of antidepressant effects. No randomized trial data were collected to assess the use of antidepressants for longer than 8 weeks.

Gabapentin, an anticonvulsant, was effective in a single large placebo-controlled trial.²² The number needed to treat (NNT) was 3.2 for the outcome of moderate or better pain relief and 13.9 for the outcome of no pain during the eighth week of treatment. The proportion of patients whose pain was much improved or who had no pain was not reported.

Controlled-release oxycodone was effective in a crossover trial in which 45% of patients had previously used opioids.²³ Tramadol may be effective but was not compared against placebo.²⁴ High-dose dextromethorphan, an N-methyl-D-aspartate (NMDA) receptor antagonist, was not shown to be effective for PHN in a small crossover trial.²⁵ Memantine, another NMDA antagonist, was also ineffective.²⁶ Acyclovir did not show any greater efficacy than placebo in a small trial.²⁷ Lorazepam and fluphenazine did not show statistically significant benefit in comparison with placebo in the amitriptyline trials.^16,18

Other therapies

Other therapies evaluated were vincristine iontophoresis, acupuncture, intrathecal methylprednisolone, and subcutaneous administration of a mixture of gangliosides (Table 1).

Iontophoresis is a process whereby topical medications are applied via electricity. Vincristine iontophoresis was no more effective than saline iontophoresis in one small trial.²⁸ Vincristine and dimethylsulfoxide iontophoresis was effective at reducing but not eliminating pain in another small trial.²⁹ Dimethylsulfoxide may have an independent analgesic effect.³⁰

Acupuncture was no more effective than mock transcutaneous electrical nerve stimulation (TENS) in 1 trial,³¹ while a smaller trial suggested a short-term effect.³²

Intrathecal methylprednisolone acetate plus lidocaine was highly effective for achieving good or excellent results (pain relief > 50%) in patients with longstanding PHN refractory to multiple conventional therapies.³³ All patients whose response to methylprednisolone was poor (8%) had had PHN for more than 5 years. The intrathecal route appears more effective than the epidural route of administration.³⁴

A mixture of gangliosides given by subcutaneous injection was more effective than placebo, but poor tolerability and derivation from bovine brain tissue severely limit its acceptability.³⁵

Sympathetic blocks using bupivacaine were more effective than intravenous lidocaine infusions in 1 trial,³⁶ but results were not reported in a fashion that conveys the proportion of patients who improved significantly.

Discussion

Effective therapies

The therapy for which evidence for efficacy is best is tricyclic antidepressants. Three placebo-controlled RCTs demonstrated that only 2 to 3 patients with PHN need to be treated to achieve 1 good or excellent result (NNT = 2-3). Since none of these studies lasted longer than 8 weeks, the long-term efficacy of antidepressants for the treatment of PHN is unknown. Follow-up of 10 patients who did well in 1 antidepressant trial found that only 2 patients were still doing well at 2 years.¹⁹

Other therapies shown to be effective in 1 or 2 trials are topical capsaicin 0.075%, gabapentin, and controlled-release oxycodone. For these studies, it is difficult to determine the number needed to treat for “meaningful” clinical benefit, although gabapentin demonstrated superiority to placebo in numerous quality-of-life measures.

For patients with severe PHN refractory to other treatments, 2 trials support benefit from intrathecal methylprednisolone and 1 trial suggests benefit from bupivacaine sympathetic blocks. Cost data for selected therapies are presented in Table 2.

TABLE 2
COSTS OF SELECTED DRUG THERAPIES

Therapies not proved effective

Therapies of uncertain benefit that have not been adequately studied in randomized controlled trials include lidocaine patch, benzydamine cream, tramadol, and vincristine (and/or dimethylsulfoxide) iontophoresis.

Therapies unlikely to be beneficial based on single trials include lorazepam, fluphenazine, dextromethorphan, memantine, acyclovir, and acupuncture. Most of the negative trials did not report power; therefore, potential benefits of these treatments cannot be excluded.

Safety and tolerability

The rates of adverse effects are high in all effective oral and topical therapies (Table 1). This situation is of special concern in elderly patients who have comorbid conditions and are taking multiple medications. Two tricyclic antidepressant trials^16,21 report a decreased incidence of side effects over time. The researchers emphasize the importance of starting at the lowest available dose with oral therapies and titrating slowly as indicated and tolerated.

No clinical complications were observed in the intrathecal steroid trials; specific side effects were not reported.

Lidocaine patch therapy has been promoted as causing clinically insignificant serum levels, no systemic side effects, and no drug–drug interactions.¹¹ However, the largest lidocaine patch RCT was too small to rule out significant but uncommon risks such as ventricular arrhythmia.¹⁰ One death that could potentially be attributed to lidocaine absorption occurred in a patient with diffuse vascular disease who was on chronic hemodialysis for renal failure. Blood lidocaine levels were not obtained because venous access was poor.

Limitations

Variations in outcomes limit comparative conclusions. Sindrup and Jensen³⁷ reviewed treatments for neuropathic pain and presented data based on a successful outcome defined as 50% reduction in pain scores, 50% pain relief, or categorical ratings of excellent, good, or moderate pain relief. Most studies used visual analog scales, but it is not clear that a 50% reduction in these measures is equivalent to clinically meaningful benefits at all levels of pain. We attempted to determine NNT data based on clearly meaningful outcomes such as “excellent or good,” “complete,” or “marked” pain relief as distinct from “moderate” or “some,” but found these data unavailable in most reports. Quality-of-life measures, such as sleep and disability ratings, may be more important than measures of pain, but were reported in only one third of the trials.

We may have failed to include relevant trials. We identified 7 potentially relevant non–English language studies. Five had no abstracts available.^38-42 One single-blind trial reported reduced pain scores with topical prostaglandin E1 dissolved in Vaseline.⁴³ One trial compared iontophoresis with lidocaine and iontophoresis with 3 different calcium channel blockers in 10 patients. The authors found that all 4 treatments reduced pain, but had not included a placebo control group.⁴⁴ We also contacted 15 content experts to identify reports of unpublished trials; none of the 6 responses received identified such reports.

The evidence base for treatment of PHN is limited. Among the RCTs reviewed, 78% enrolled 50 or fewer patients. Because most of the subjects had PHN lasting longer than 1 year, our conclusions may not apply to patients with PHN of shorter duration. The latter group represents the majority of subjects with PHN presenting to primary care physicians.

We used the Jadad scale⁷ as an attempt to quantitatively assess the methodologic quality of the studies we reviewed. In general, explicit validity checklists with summary scores have not consistently been shown to provide more reliable assessments of validity than qualitative assessments.^45-4 The Jadad scale is the first validity checklist that has some rigorous evidence supporting its use,^7,48 although its inter-rater reliability has been questioned.⁴⁹ We found the Jadad scale had an inherent bias against therapies that could not be adequately double-blinded. Thus 2 trials with a score of 1 were included.^29,36 In 1 trial of vincristine iontophoresis, the authors described the trial as single-blinded and provided ample explanation of why double-blinding was not achieved.²⁹ In 1 trial of sympathetic blocks, the treatment studied included an invasive procedure; therefore, there was no apparent way for the procedure itself to be double-blinded.³⁶

The Jadad scale does not account for some threats to validity of included studies. We encountered numerous methodologic flaws, such as lack of intention to treat analysis in parallel trials (11) and lack of washout periods in crossover trials (4). Further limitations to interpretation of selected study results included potentially significant baseline differences between groups (8), small numbers, and short study durations. A list of the studies with these specific methodologic concerns is available in Table W3.

Conclusions

For patients with PHN lasting less than 6 months, spontaneous resolution is common and treatment decisions are largely empiric and not evidence based. For PHN of longer duration, treatments shown to be more effective than placebo include tricyclic antidepressants, topical capsaicin 0.075%, gabapentin, and controlled-release oxycodone. These treatments all have adverse effects or costs that need to be considered on an individual basis. Lidocaine patch therapy may be safer for most patients but may be no more effective than placebo and is not suitable for patients with trigeminal PHN. Patients with PHN refractory to the currently available and studied topical and oral agents should be considered for intrathecal steroid therapy.

Acknowledgments

The authors wish to thank Susan Meadows, MLS, Susan Elliott, MLS, Stacey Rautzhan, and Steve Calloway, RPh, for their assistance and Sigurdur Helgason, MD, Carin Reust, MD, Steven Zweig, MD, MSPH, and Alan Adelman, MD, MS, for editorial review.

Postherpetic neuralgia (PHN), the most common complication of herpes zoster, is much more prevalent among adults older than 50 years than in younger people.^1,2 The largest English-language prospective study of patients presenting with zoster suggests that the average family physician can expect to see 4 cases of zoster per year and 1 case of PHN lasting more than 3 months every 3 years.³ Among placebo cohorts from randomized controlled trials (RCTs) of acute zoster treatment, the incidence of pain at 3 months has been reported as 17% to 60%; at 6 months, 5% to 39%.⁴ There is limited evidence that therapies for acute zoster have an impact on PHN.⁴

This review addresses therapies to reduce pain or improve quality of life in patients with PHN. The condition has been variously defined in terms of timing (either following resolution of acute zoster [rash healing] or a defined time after onset of zoster), duration (any time after zoster or a minimum of 6 months after zoster), and type of pain (such as lancinating pain or allodynia [pain caused by a stimulus that does not nor mally provoke pain]).^5,6 PHN may include a spectrum of presentations, from brief intermittent mild pain that resolves spontaneously to chronic persistent disabling pain recalcitrant to multiple therapies. To avoid missing potentially relevant findings, we defined PHN broadly as any pain after cutaneous healing of zoster.

Methods

Search strategy

Medline (1966 to present) was searched on October 19, 2000. The search combined the terms “post-herpetic or postherpetic” and “neuralgia or neuropathy or pain” and publication type “clinical trial (including phases I to IV) or controlled clinical trial or randomized controlled trial.” The Cochrane Controlled Trials Registry 2000, Issue 3, was searched with the same terms. Current Contents was searched to identify more recent references. We also identified trials through article reference lists (from included trials and 40 reviews), contact of authors and content experts, and the Food and Drug Administration (FDA) Web site.

Selection criteria

Inclusion criteria for this review were RCTs that enrolled patients with PHN (history of zoster, pain in the dermatomal distribution of the zoster rash, and pain persisting or occurring after resolution of the zoster rash), addressed relevant end points (pain resolution, pain severity, effect on quality of life), and had full reports available in English. Since responses to initial therapy may change over time, we included only trials with evaluation periods lasting more than 24 hours.

The authors independently evaluated trials meeting these inclusion criteria for quality and independently extracted data. Quality was evaluated using the Jadad scale,⁷ which addresses selected criteria (randomization technique, allocation concealment, blinding, accounting of dropouts) and rates methodologic quality on a 5-point scale, with 5 representing the highest score. Differences were resolved through discussion. Trials scoring only 1 point were excluded except for 2 instances noted in our discussion.

Results

Searching identified 186 potential trials, of which 92 were excluded as irrelevant on the basis of titles and abstracts alone. Of the 94 citations reviewed in greater detail, 64 were excluded for the following reasons: not describing a trial (10), not describing a trial of treatment for PHN (10), describing an uncontrolled trial (7), no randomization (7), evaluation period limited to 24 hours or less (13), duplicate publication (4), language other than English (7), not providing results specifically for patients with PHN (3), and available only in abstract form (3). One unpublished trial of mexiletine was identified through a review by Hanania and Brietstein; Dr Hanania informed us, however, that this trial had been stopped early because the treatment group had experienced serious side effects. One controlled trial was excluded because correspondence with the investigator did not confirm that it had been randomized. One trial was excluded because it included only 6 patients with PHN. (A list of excluded studies is available in Table W1.)

Of the remaining 27 trials reviewed for methodologic quality, most (16) received a Jadad score of 4. The 2 authors had substantial agreement on quality ratings ( = 0.75). Table W2 provides details of treatment regimens, quality ratings, ages of subjects, and duration of PHN. One trial with a Jadad score of 1 was excluded.⁸ Most subjects were elderly and had had PHN for longer than 6 months.

Topical therapies

Topical therapies evaluated were lidocaine, capsaicin, and benzydamine (Table 1). Lidocaine patch therapy is the only agent with a specific FDA indication for PHN. We found few trials supporting the FDA approval. The only published RCT of relatively unselected patients with PHN (n = 35) showed significant benefit versus placebo but was excluded because evaluation sessions had been limited to 12 hours.⁹ We reviewed a report of an unpublished RCT comparing lidocaine patch with vehicle placebo used in the application for FDA approval.¹⁰ This trial found a large, statistically significant reduction in pain scores with placebo throughout the 3- to 4-week trial. This trial found a similar statistically significant reduction in pain scores with lidocaine patch and no significant difference comparing lidocaine patch with placebo.

Three findings in the unpublished trial were used to support arguments for efficacy: (1) a statistically significant difference in the pain relief score at the final visit; (2) differences in allodynia (based on investigators’ sensory skin testing, described as stroking the maximally painful area with a foam brush and recording the pain scale rating) at the beginning of the trial; and (3) a greater increase in pain scores among lidocaine subjects upon trial conclusion (ie, after stopping study medication). The clinical relevance of these 3 findings is unclear.

The FDA declined to approve lidocaine patch therapy on the basis of these 2 studies and required an additional trial to demonstrate benefit. An “enriched enrollment study” involved subjects who had used lidocaine patch for at least 1 month and received at least moderate relief but had pain without the patch.¹¹ Lidocaine patch was clearly effective in this highly selected cohort.

Capsaicin 0.075% cream was effective in 2 trials of patients with severe refractory PHN.^12,13 The benefit appeared modest in the larger trial (pain was eliminated or nearly eliminated in less than 20% of capsaicin patients) and greater in the smaller trial. Blinding had limited efficacy because of the stinging effect of capsaicin.

Benzydamine cream, an antiprostaglandin, was not effective in a 2-week crossover trial.¹⁴ The cream showed a nonsignificant trend for pain reduction in an earlier 2-week crossover trial.¹⁵

TABLE 1

Oral therapies

Oral therapies evaluated were tricyclic antidepressants, gabapentin, oxycodone, tramadol, dextromethorphan, memantine, acyclovir, lorazepam, and fluphenazine (Table 1).

Tricyclic antidepressants have been shown to be effective in multiple small short-term crossover trials. Amitriptyline was highly effective in 2 placebo-controlled trials.^16,17 In 1 of these trials, amitriptyline was more effective than lorazepam.¹⁶ In another trial, amitriptyline was more effective than fluphenazine (a phenothiazine) and glycopyrrolate placebo.¹⁸ Nortriptyline was as effective as amitriptyline in a comparison trial,¹⁹ while maprotiline was not.²⁰ Desipramine was highly effective in a trial using benztropine as an “active placebo” in that the anticholinergic properties of benztropine were used to match the side effects of desipramine.²¹ In all these studies, the analgesic effects of tricyclic antidepressants appeared independent of antidepressant effects. No randomized trial data were collected to assess the use of antidepressants for longer than 8 weeks.

Gabapentin, an anticonvulsant, was effective in a single large placebo-controlled trial.²² The number needed to treat (NNT) was 3.2 for the outcome of moderate or better pain relief and 13.9 for the outcome of no pain during the eighth week of treatment. The proportion of patients whose pain was much improved or who had no pain was not reported.

Controlled-release oxycodone was effective in a crossover trial in which 45% of patients had previously used opioids.²³ Tramadol may be effective but was not compared against placebo.²⁴ High-dose dextromethorphan, an N-methyl-D-aspartate (NMDA) receptor antagonist, was not shown to be effective for PHN in a small crossover trial.²⁵ Memantine, another NMDA antagonist, was also ineffective.²⁶ Acyclovir did not show any greater efficacy than placebo in a small trial.²⁷ Lorazepam and fluphenazine did not show statistically significant benefit in comparison with placebo in the amitriptyline trials.^16,18

Other therapies

Other therapies evaluated were vincristine iontophoresis, acupuncture, intrathecal methylprednisolone, and subcutaneous administration of a mixture of gangliosides (Table 1).

Iontophoresis is a process whereby topical medications are applied via electricity. Vincristine iontophoresis was no more effective than saline iontophoresis in one small trial.²⁸ Vincristine and dimethylsulfoxide iontophoresis was effective at reducing but not eliminating pain in another small trial.²⁹ Dimethylsulfoxide may have an independent analgesic effect.³⁰

Acupuncture was no more effective than mock transcutaneous electrical nerve stimulation (TENS) in 1 trial,³¹ while a smaller trial suggested a short-term effect.³²

Intrathecal methylprednisolone acetate plus lidocaine was highly effective for achieving good or excellent results (pain relief > 50%) in patients with longstanding PHN refractory to multiple conventional therapies.³³ All patients whose response to methylprednisolone was poor (8%) had had PHN for more than 5 years. The intrathecal route appears more effective than the epidural route of administration.³⁴

A mixture of gangliosides given by subcutaneous injection was more effective than placebo, but poor tolerability and derivation from bovine brain tissue severely limit its acceptability.³⁵

Sympathetic blocks using bupivacaine were more effective than intravenous lidocaine infusions in 1 trial,³⁶ but results were not reported in a fashion that conveys the proportion of patients who improved significantly.

Discussion

Effective therapies

The therapy for which evidence for efficacy is best is tricyclic antidepressants. Three placebo-controlled RCTs demonstrated that only 2 to 3 patients with PHN need to be treated to achieve 1 good or excellent result (NNT = 2-3). Since none of these studies lasted longer than 8 weeks, the long-term efficacy of antidepressants for the treatment of PHN is unknown. Follow-up of 10 patients who did well in 1 antidepressant trial found that only 2 patients were still doing well at 2 years.¹⁹

Other therapies shown to be effective in 1 or 2 trials are topical capsaicin 0.075%, gabapentin, and controlled-release oxycodone. For these studies, it is difficult to determine the number needed to treat for “meaningful” clinical benefit, although gabapentin demonstrated superiority to placebo in numerous quality-of-life measures.

For patients with severe PHN refractory to other treatments, 2 trials support benefit from intrathecal methylprednisolone and 1 trial suggests benefit from bupivacaine sympathetic blocks. Cost data for selected therapies are presented in Table 2.

TABLE 2
COSTS OF SELECTED DRUG THERAPIES

Therapies not proved effective

Therapies of uncertain benefit that have not been adequately studied in randomized controlled trials include lidocaine patch, benzydamine cream, tramadol, and vincristine (and/or dimethylsulfoxide) iontophoresis.

Therapies unlikely to be beneficial based on single trials include lorazepam, fluphenazine, dextromethorphan, memantine, acyclovir, and acupuncture. Most of the negative trials did not report power; therefore, potential benefits of these treatments cannot be excluded.

Safety and tolerability

The rates of adverse effects are high in all effective oral and topical therapies (Table 1). This situation is of special concern in elderly patients who have comorbid conditions and are taking multiple medications. Two tricyclic antidepressant trials^16,21 report a decreased incidence of side effects over time. The researchers emphasize the importance of starting at the lowest available dose with oral therapies and titrating slowly as indicated and tolerated.

No clinical complications were observed in the intrathecal steroid trials; specific side effects were not reported.

Lidocaine patch therapy has been promoted as causing clinically insignificant serum levels, no systemic side effects, and no drug–drug interactions.¹¹ However, the largest lidocaine patch RCT was too small to rule out significant but uncommon risks such as ventricular arrhythmia.¹⁰ One death that could potentially be attributed to lidocaine absorption occurred in a patient with diffuse vascular disease who was on chronic hemodialysis for renal failure. Blood lidocaine levels were not obtained because venous access was poor.

Limitations

Variations in outcomes limit comparative conclusions. Sindrup and Jensen³⁷ reviewed treatments for neuropathic pain and presented data based on a successful outcome defined as 50% reduction in pain scores, 50% pain relief, or categorical ratings of excellent, good, or moderate pain relief. Most studies used visual analog scales, but it is not clear that a 50% reduction in these measures is equivalent to clinically meaningful benefits at all levels of pain. We attempted to determine NNT data based on clearly meaningful outcomes such as “excellent or good,” “complete,” or “marked” pain relief as distinct from “moderate” or “some,” but found these data unavailable in most reports. Quality-of-life measures, such as sleep and disability ratings, may be more important than measures of pain, but were reported in only one third of the trials.

We may have failed to include relevant trials. We identified 7 potentially relevant non–English language studies. Five had no abstracts available.^38-42 One single-blind trial reported reduced pain scores with topical prostaglandin E1 dissolved in Vaseline.⁴³ One trial compared iontophoresis with lidocaine and iontophoresis with 3 different calcium channel blockers in 10 patients. The authors found that all 4 treatments reduced pain, but had not included a placebo control group.⁴⁴ We also contacted 15 content experts to identify reports of unpublished trials; none of the 6 responses received identified such reports.

The evidence base for treatment of PHN is limited. Among the RCTs reviewed, 78% enrolled 50 or fewer patients. Because most of the subjects had PHN lasting longer than 1 year, our conclusions may not apply to patients with PHN of shorter duration. The latter group represents the majority of subjects with PHN presenting to primary care physicians.

We used the Jadad scale⁷ as an attempt to quantitatively assess the methodologic quality of the studies we reviewed. In general, explicit validity checklists with summary scores have not consistently been shown to provide more reliable assessments of validity than qualitative assessments.^45-4 The Jadad scale is the first validity checklist that has some rigorous evidence supporting its use,^7,48 although its inter-rater reliability has been questioned.⁴⁹ We found the Jadad scale had an inherent bias against therapies that could not be adequately double-blinded. Thus 2 trials with a score of 1 were included.^29,36 In 1 trial of vincristine iontophoresis, the authors described the trial as single-blinded and provided ample explanation of why double-blinding was not achieved.²⁹ In 1 trial of sympathetic blocks, the treatment studied included an invasive procedure; therefore, there was no apparent way for the procedure itself to be double-blinded.³⁶

The Jadad scale does not account for some threats to validity of included studies. We encountered numerous methodologic flaws, such as lack of intention to treat analysis in parallel trials (11) and lack of washout periods in crossover trials (4). Further limitations to interpretation of selected study results included potentially significant baseline differences between groups (8), small numbers, and short study durations. A list of the studies with these specific methodologic concerns is available in Table W3.

Conclusions

For patients with PHN lasting less than 6 months, spontaneous resolution is common and treatment decisions are largely empiric and not evidence based. For PHN of longer duration, treatments shown to be more effective than placebo include tricyclic antidepressants, topical capsaicin 0.075%, gabapentin, and controlled-release oxycodone. These treatments all have adverse effects or costs that need to be considered on an individual basis. Lidocaine patch therapy may be safer for most patients but may be no more effective than placebo and is not suitable for patients with trigeminal PHN. Patients with PHN refractory to the currently available and studied topical and oral agents should be considered for intrathecal steroid therapy.

Acknowledgments

The authors wish to thank Susan Meadows, MLS, Susan Elliott, MLS, Stacey Rautzhan, and Steve Calloway, RPh, for their assistance and Sigurdur Helgason, MD, Carin Reust, MD, Steven Zweig, MD, MSPH, and Alan Adelman, MD, MS, for editorial review.

Among underserved populations, the burden of tobacco is substantial. There is a clear association between poverty and high rates of tobacco use,^1-3 and smoking is more prevalent among the uninsured (39%) than those with insurance (23%).⁴ Smoking cessation interventions can be successful among low-income and minority patients, especially when tailored to these populations.^5-8 Tobacco counseling, including simple advice to quit, has been shown effective in primary care.^9-11 Since disadvantaged patients, including 63% of the uninsured,¹² are commonly seen in primary care settings, primary care providers are in a unique position to impact tobacco use in underserved patients.

Previous research on cessation counseling rates in low-income patients has yielded conflicting results. Taira and colleagues¹¹ demonstrated that cessation advice by primary care providers was given more frequently to low-income groups. However, this study’s results were based on a written patient questionnaire, and recall may have been a significant limitation. Another study examined physician-reported rates of tobacco cessation counseling, and found that cessation was addressed more frequently with health maintenance organization (HMO)–insured patients (30%) than Medicaid patients (24%).¹³ However, this analysis did not differentiate between primary care providers and specialists, and neither of these studies identified low-income uninsured patients.

It thus remains unclear whether this effective intervention is routinely provided to underserved patients, including the uninsured, in primary care settings. Using a provider survey instrument that clearly identified medically indigent patients, this study examined the frequency with which primary care providers address tobacco use with their Medicaid-insured and uninsured patients compared with those with private or HMO insurance.

Methods

This study was conducted in the 7 primary care practices in the Colorado Research Network (CaReNet) in 1998 and 1999. CaReNet is a state-wide primary care, practice-based research network founded in 1997 with a particular focus on disadvantaged populations, including rural people, minorities, and the urban poor. The practices in CaReNet are affiliated with the University of Colorado Department of Family Medicine. Of the 7 practices, 4 are family medicine residency sites, 2 are federally-funded community health centers, and 1 is a clinic for the medically indigent. The provider mix in CaReNet includes 56% residents (residents average approximately 3 half-day clinics weekly), 21% full-time clinical faculty, 7% private physicians, and 15% other providers (nurse practitioners, physician assistants, and so forth). At the time of our study, none of the practices had a comprehensive tobacco cessation program on site. Colorado Medicaid recipients were eligible for a limited amount of smoking cessation products (this benefit required prior authorization), but Medicaid did not cover comprehensive programs.

A modified version of the 1994 National Ambulatory Medical Care Survey (NAMCS) was administered in each CaReNet practice. The NAMCS instrument is a physician survey that collects information about an ambulatory visit; it has been used by the National Center for Health Statistics since 1973 to analyze trends of ambulatory care. In the context of our study, the key modification was the addition of “uninsured” in the Expected Source of Payment category. This category included patients who were in 1 of several programs that discount charges on the basis of income, thus covering some of the costs of care. All providers received detailed instructions on completing this modified NAMCS form.

Each CaReNet practice collected data on a total of 400 patient visits in 1-week cycles (100 patients per cycle), quarterly, for 1 year. We used the typical NAMCS protocol of collecting data on every second patient presenting for medical care during the study period.¹⁴ The anonymous visit survey forms were coded using standard NAMCS nomenclature. Only patients aged between 13 years and 65 years were included in this analysis because there are almost no uninsured people older than 65 years. To identify patients with private insurance, the options “Private/commercial” and “HMO/other prepaid” were combined (hereafter referred to as “Private/HMO”).

For the present study, we examined the impact of patient insurance on 2 primary outcomes: (1) patient smoking status, and (2) whether smokers received smoking cessation counseling. Each provider coded smoking status as “Yes,” “No,” or “Unknown.” Only patients with a known smoking status (90% of sample) were included in the present analysis. For those patients coded as smokers, we determined whether providers checked the “Smoking Cessation” box.

Analysis

To examine the association between insurance group and study outcomes, we used chi-square tests to determine whether insurance group and other patient demographics (sex, age, ethnicity, and race) were reliably associated with smoking status and cessation counseling. Next, for each primary outcome, we conducted multivariate analyses to examine the effect of patient insurance, while controlling for other important demographic factors (ie, those with P values 0.20 in univariate analyses,¹⁵ as well as additional factors that may account for variability in this relationship. These factors included duration of visit, whether the patient had been seen before in the practice, and whether the patient had at least 1 of the chronic conditions listed on the NAMCS form (hypertension, depression, obesity, or hypercholesterolemia). Because initial random effects analyses revealed no significant practice site effects on the frequency of tobacco use and cessation counseling, all analyses include patient-level data.

The Colorado Multiple Institutional Review Board approved our study design.

Results

Description of sample

CaReNet providers completed NAMCS forms on 2773 patient encounters of 2800 eligible visits (99% completion rate). For this study, of the 2773 encounters, 1443 remained after excluding patients younger than 13 or older than 65 years, and those with sources of payment other than Medicaid, Uninsured, or Private/HMO. As shown in (Table 1), CaReNet patients in the present study were demographically diverse, with a high percentage who were Hispanic (26%), female (74%), or low-income (39% uninsured, 22% Medicaid).

TABLE 1
DEMOGRAPHIC CHARACTERISTICS OF CARENET STUDY SAMPLE

Univariate and multivariate analysis of smoking

A total of 351 patients in the study sample (24%) were identified as smokers. As expected, smoking was significantly more prevalent in the Medicaid and uninsured groups (Table W1*).

(Table 2) presents multivariate logistic regression results showing the significant relationship between insurance and smoking status after controlling for other important demographic and practice variables. Uninsured patients had similar rates of smoking as those with Medicaid; however, smoking among Private/HMO–insured patients was approximately half as frequent as among the uninsured.

In addition to patient insurance, ethnicity and clinical factors predicted whether patients smoked. Non-Hispanic patients were more than twice as likely to be identified as smokers compared with Hispanic patients (P <.001). Also, patients who were new to the practice or who had at least one chronic condition were significantly more likely to be identified as smokers (P = .011 and P = .001, respectively).

Table 2
LOGISTIC REGRESSION RESULTS: RELATIONSHIP OF PATIENT FACTORS WITH LIKELIHOOD OF SMOKING

Univeriate and multivariate analysis of cessation advice or counseling

The second primary analysis examined whether insurance is associated with how often smokers are counseled during visits. Out of 351 smokers, 129 (37%) received tobacco counseling during the medical encounter. Private/HMO insurance and duration of visit were the only factors univariately associated with whether a smoker received counseling (Table W2*).

Multivariate results indicate that patient insurance remained the only significant variable after controlling for other factors that might explain whether smokers received counseling. Smokers with Medicaid were more than twice as likely, and Private/HMO–insured smokers were more than 3 times as likely as uninsured patients (P <.001) to receive smoking cessation counseling (Table 3).

TABLE 3
LOGISTIC REGRESSION RESULTS: PATIENT FACTORS ASSOCIATED WITH LIKELIHOOD OF RECEIVING SMOKING CESSATION COUNSELING

Discussion

These findings demonstrate that although smoking is more common in CaReNet’s Medicaid and uninsured patients, providers gave cessation advice less often to these patients. The actual prevalence of tobacco use may be even greater than we think because providers may underreport it, but our results are similar to national trends.⁴ The decreased rate of tobacco counseling in underserved patients is in contrast to the findings in a study that were based on patient recall,¹¹ rather than the provider-report methodology of NAMCS. However, our counseling results are consistent with a national NAMCS analysis, which found that tobacco use was addressed more frequently with HMO-insured patients than Medicaid patients.¹³ In that study, the overall primary care counseling rate (33%) was similar to that of CaReNet providers (37%). To the best of our knowledge, our finding of a lower rate of tobacco counseling in uninsured patients has not been previously reported.

Our study does not address why providers are less likely to advise Medicaid or uninsured patients to quit smoking. It is possible that tobacco interventions, such as pharmacologic aids and comprehensive cessation programs, may not be available to these groups because of cost. Providers may simply be reflecting this situation by not addressing cessation. Even so, cost and access barriers do not explain why providers would be less likely to give simple cessation advice to disadvantaged smokers. One possibility is that these findings may indicate a lower quality of care for these patients. Other preventive care measures have been shown to be performed less often in uninsured patients,¹⁶ and several studies have documented a lower quality of care for Medicaid and uninsured patients with chronic diseases.^17-19

Limitations

A major limitation of our study is that the uninsured or Medicaid groups may have included sicker or more complex patients at the surveyed visits, thus there may have been less time to devote to tobacco cessation advice during that clinic visit. Unfortunately, the NAMCS instrument does not readily measure disease severity or case mix. In our analysis, we controlled for the presence of 1 or more chronic diseases (limited in NAMCS to 4 specific conditions), but this is only a crude measure of patient complexity. If patients in one of the payment groups were sicker, they might have had more frequent clinic visits, and tobacco cessation may have been addressed at higher rates over time than were found in this cross-sectional study. However, even in the presence of major morbidities, the uninsured often lack continuity because of their tenuous access to care.

If the payer mix of residents and faculty was significantly different, and residents addressed tobacco use at a different rate than faculty, this could explain some of the counseling differences. Unfortunately, this NAMCS instrument is anonymous and cannot identify the type of provider. Similarly, it is possible that the type of visit (acute care, chronic care, or prevention) may account for some of the findings. However, NAMCS also does not specify type of visit and there may be considerable overlap at any given encounter.

Our study administered NAMCS to the practices that make up CaReNet, and the results are not necessarily generalizable to other populations. There is substantial regional variation in health care access programs for the uninsured²⁰; therefore the uninsured patients in CaReNet may not be representative of uninsured in primary care elsewhere. Also, the demographics of CaReNet include higher percentages of Hispanics and Medicaid recipients compared with a national analysis of primary care trends.²¹ CaReNet more closely resembles community health centers,²² except CaReNet has a greater number of Hispanic patients and fewer black patients, reflecting the particular demographics of Colorado. However, the smoking prevalence rates we found in the privately insured, Medicaid, and uninsured groups were similar to national patterns.

Conclusions

Our study argues for the inclusion of a separate payment category that clearly identifies the uninsured in NAMCS and other data collection instruments. Future studies on tobacco counseling rates should be designed to differentiate factors associated with the lower rate of counseling in disadvantaged populations, such as patient complexity, competing demands, lack of access to cessation resources, or lower standards of care. Identification of these factors may be valuable in implementing interventions to improve the rate of counseling for these patients.

If national tobacco goals are to be realized, then socioeconomic disparities in counseling need to be addressed. Our results show that primary care providers can substantially improve the tobacco counseling rate among disadvantaged smokers. As this occurs, the rate of smoking in these patients can be expected to decrease.

*Tables W1 and W2 are available on the JFP Web site, www.jfponline.com.

Acknowledgments

We appreciate the financial support of CaReNet by the University of Colorado School of Medicine Academic Enrichment Fund. We would also like to thank the faculty, residents, and staff at the following CaReNet sites for their assistance with this study: CU Care, Denver, Colorado; St. Mary’s Family Practice, Grand Junction, Colorado; Brighton Salud Family Health Center, Brighton, Colorado; Rose Family Medicine Center, Denver, Colorado; Swedish Family Medicine Center, Littleton, Colorado; AF Williams Family Medicine Center, Denver, Colorado; and La CasaQuigg Newton Health Center, Denver, Colorado. We are also grateful to Elizabeth Staton and Michael Huiras, MD, for their comments on the manuscript. The authors deny any conflict of interest.

Among underserved populations, the burden of tobacco is substantial. There is a clear association between poverty and high rates of tobacco use,^1-3 and smoking is more prevalent among the uninsured (39%) than those with insurance (23%).⁴ Smoking cessation interventions can be successful among low-income and minority patients, especially when tailored to these populations.^5-8 Tobacco counseling, including simple advice to quit, has been shown effective in primary care.^9-11 Since disadvantaged patients, including 63% of the uninsured,¹² are commonly seen in primary care settings, primary care providers are in a unique position to impact tobacco use in underserved patients.

Previous research on cessation counseling rates in low-income patients has yielded conflicting results. Taira and colleagues¹¹ demonstrated that cessation advice by primary care providers was given more frequently to low-income groups. However, this study’s results were based on a written patient questionnaire, and recall may have been a significant limitation. Another study examined physician-reported rates of tobacco cessation counseling, and found that cessation was addressed more frequently with health maintenance organization (HMO)–insured patients (30%) than Medicaid patients (24%).¹³ However, this analysis did not differentiate between primary care providers and specialists, and neither of these studies identified low-income uninsured patients.

It thus remains unclear whether this effective intervention is routinely provided to underserved patients, including the uninsured, in primary care settings. Using a provider survey instrument that clearly identified medically indigent patients, this study examined the frequency with which primary care providers address tobacco use with their Medicaid-insured and uninsured patients compared with those with private or HMO insurance.

Methods

This study was conducted in the 7 primary care practices in the Colorado Research Network (CaReNet) in 1998 and 1999. CaReNet is a state-wide primary care, practice-based research network founded in 1997 with a particular focus on disadvantaged populations, including rural people, minorities, and the urban poor. The practices in CaReNet are affiliated with the University of Colorado Department of Family Medicine. Of the 7 practices, 4 are family medicine residency sites, 2 are federally-funded community health centers, and 1 is a clinic for the medically indigent. The provider mix in CaReNet includes 56% residents (residents average approximately 3 half-day clinics weekly), 21% full-time clinical faculty, 7% private physicians, and 15% other providers (nurse practitioners, physician assistants, and so forth). At the time of our study, none of the practices had a comprehensive tobacco cessation program on site. Colorado Medicaid recipients were eligible for a limited amount of smoking cessation products (this benefit required prior authorization), but Medicaid did not cover comprehensive programs.

A modified version of the 1994 National Ambulatory Medical Care Survey (NAMCS) was administered in each CaReNet practice. The NAMCS instrument is a physician survey that collects information about an ambulatory visit; it has been used by the National Center for Health Statistics since 1973 to analyze trends of ambulatory care. In the context of our study, the key modification was the addition of “uninsured” in the Expected Source of Payment category. This category included patients who were in 1 of several programs that discount charges on the basis of income, thus covering some of the costs of care. All providers received detailed instructions on completing this modified NAMCS form.

Each CaReNet practice collected data on a total of 400 patient visits in 1-week cycles (100 patients per cycle), quarterly, for 1 year. We used the typical NAMCS protocol of collecting data on every second patient presenting for medical care during the study period.¹⁴ The anonymous visit survey forms were coded using standard NAMCS nomenclature. Only patients aged between 13 years and 65 years were included in this analysis because there are almost no uninsured people older than 65 years. To identify patients with private insurance, the options “Private/commercial” and “HMO/other prepaid” were combined (hereafter referred to as “Private/HMO”).

For the present study, we examined the impact of patient insurance on 2 primary outcomes: (1) patient smoking status, and (2) whether smokers received smoking cessation counseling. Each provider coded smoking status as “Yes,” “No,” or “Unknown.” Only patients with a known smoking status (90% of sample) were included in the present analysis. For those patients coded as smokers, we determined whether providers checked the “Smoking Cessation” box.

Analysis

To examine the association between insurance group and study outcomes, we used chi-square tests to determine whether insurance group and other patient demographics (sex, age, ethnicity, and race) were reliably associated with smoking status and cessation counseling. Next, for each primary outcome, we conducted multivariate analyses to examine the effect of patient insurance, while controlling for other important demographic factors (ie, those with P values 0.20 in univariate analyses,¹⁵ as well as additional factors that may account for variability in this relationship. These factors included duration of visit, whether the patient had been seen before in the practice, and whether the patient had at least 1 of the chronic conditions listed on the NAMCS form (hypertension, depression, obesity, or hypercholesterolemia). Because initial random effects analyses revealed no significant practice site effects on the frequency of tobacco use and cessation counseling, all analyses include patient-level data.

The Colorado Multiple Institutional Review Board approved our study design.

Results

Description of sample

CaReNet providers completed NAMCS forms on 2773 patient encounters of 2800 eligible visits (99% completion rate). For this study, of the 2773 encounters, 1443 remained after excluding patients younger than 13 or older than 65 years, and those with sources of payment other than Medicaid, Uninsured, or Private/HMO. As shown in (Table 1), CaReNet patients in the present study were demographically diverse, with a high percentage who were Hispanic (26%), female (74%), or low-income (39% uninsured, 22% Medicaid).

TABLE 1
DEMOGRAPHIC CHARACTERISTICS OF CARENET STUDY SAMPLE

Univariate and multivariate analysis of smoking

A total of 351 patients in the study sample (24%) were identified as smokers. As expected, smoking was significantly more prevalent in the Medicaid and uninsured groups (Table W1*).

(Table 2) presents multivariate logistic regression results showing the significant relationship between insurance and smoking status after controlling for other important demographic and practice variables. Uninsured patients had similar rates of smoking as those with Medicaid; however, smoking among Private/HMO–insured patients was approximately half as frequent as among the uninsured.

In addition to patient insurance, ethnicity and clinical factors predicted whether patients smoked. Non-Hispanic patients were more than twice as likely to be identified as smokers compared with Hispanic patients (P <.001). Also, patients who were new to the practice or who had at least one chronic condition were significantly more likely to be identified as smokers (P = .011 and P = .001, respectively).

Table 2
LOGISTIC REGRESSION RESULTS: RELATIONSHIP OF PATIENT FACTORS WITH LIKELIHOOD OF SMOKING

Univeriate and multivariate analysis of cessation advice or counseling

The second primary analysis examined whether insurance is associated with how often smokers are counseled during visits. Out of 351 smokers, 129 (37%) received tobacco counseling during the medical encounter. Private/HMO insurance and duration of visit were the only factors univariately associated with whether a smoker received counseling (Table W2*).

Multivariate results indicate that patient insurance remained the only significant variable after controlling for other factors that might explain whether smokers received counseling. Smokers with Medicaid were more than twice as likely, and Private/HMO–insured smokers were more than 3 times as likely as uninsured patients (P <.001) to receive smoking cessation counseling (Table 3).

TABLE 3
LOGISTIC REGRESSION RESULTS: PATIENT FACTORS ASSOCIATED WITH LIKELIHOOD OF RECEIVING SMOKING CESSATION COUNSELING

Discussion

These findings demonstrate that although smoking is more common in CaReNet’s Medicaid and uninsured patients, providers gave cessation advice less often to these patients. The actual prevalence of tobacco use may be even greater than we think because providers may underreport it, but our results are similar to national trends.⁴ The decreased rate of tobacco counseling in underserved patients is in contrast to the findings in a study that were based on patient recall,¹¹ rather than the provider-report methodology of NAMCS. However, our counseling results are consistent with a national NAMCS analysis, which found that tobacco use was addressed more frequently with HMO-insured patients than Medicaid patients.¹³ In that study, the overall primary care counseling rate (33%) was similar to that of CaReNet providers (37%). To the best of our knowledge, our finding of a lower rate of tobacco counseling in uninsured patients has not been previously reported.

Our study does not address why providers are less likely to advise Medicaid or uninsured patients to quit smoking. It is possible that tobacco interventions, such as pharmacologic aids and comprehensive cessation programs, may not be available to these groups because of cost. Providers may simply be reflecting this situation by not addressing cessation. Even so, cost and access barriers do not explain why providers would be less likely to give simple cessation advice to disadvantaged smokers. One possibility is that these findings may indicate a lower quality of care for these patients. Other preventive care measures have been shown to be performed less often in uninsured patients,¹⁶ and several studies have documented a lower quality of care for Medicaid and uninsured patients with chronic diseases.^17-19

Limitations

A major limitation of our study is that the uninsured or Medicaid groups may have included sicker or more complex patients at the surveyed visits, thus there may have been less time to devote to tobacco cessation advice during that clinic visit. Unfortunately, the NAMCS instrument does not readily measure disease severity or case mix. In our analysis, we controlled for the presence of 1 or more chronic diseases (limited in NAMCS to 4 specific conditions), but this is only a crude measure of patient complexity. If patients in one of the payment groups were sicker, they might have had more frequent clinic visits, and tobacco cessation may have been addressed at higher rates over time than were found in this cross-sectional study. However, even in the presence of major morbidities, the uninsured often lack continuity because of their tenuous access to care.

If the payer mix of residents and faculty was significantly different, and residents addressed tobacco use at a different rate than faculty, this could explain some of the counseling differences. Unfortunately, this NAMCS instrument is anonymous and cannot identify the type of provider. Similarly, it is possible that the type of visit (acute care, chronic care, or prevention) may account for some of the findings. However, NAMCS also does not specify type of visit and there may be considerable overlap at any given encounter.

Our study administered NAMCS to the practices that make up CaReNet, and the results are not necessarily generalizable to other populations. There is substantial regional variation in health care access programs for the uninsured²⁰; therefore the uninsured patients in CaReNet may not be representative of uninsured in primary care elsewhere. Also, the demographics of CaReNet include higher percentages of Hispanics and Medicaid recipients compared with a national analysis of primary care trends.²¹ CaReNet more closely resembles community health centers,²² except CaReNet has a greater number of Hispanic patients and fewer black patients, reflecting the particular demographics of Colorado. However, the smoking prevalence rates we found in the privately insured, Medicaid, and uninsured groups were similar to national patterns.

Conclusions

Our study argues for the inclusion of a separate payment category that clearly identifies the uninsured in NAMCS and other data collection instruments. Future studies on tobacco counseling rates should be designed to differentiate factors associated with the lower rate of counseling in disadvantaged populations, such as patient complexity, competing demands, lack of access to cessation resources, or lower standards of care. Identification of these factors may be valuable in implementing interventions to improve the rate of counseling for these patients.

If national tobacco goals are to be realized, then socioeconomic disparities in counseling need to be addressed. Our results show that primary care providers can substantially improve the tobacco counseling rate among disadvantaged smokers. As this occurs, the rate of smoking in these patients can be expected to decrease.

*Tables W1 and W2 are available on the JFP Web site, www.jfponline.com.

Acknowledgments

We appreciate the financial support of CaReNet by the University of Colorado School of Medicine Academic Enrichment Fund. We would also like to thank the faculty, residents, and staff at the following CaReNet sites for their assistance with this study: CU Care, Denver, Colorado; St. Mary’s Family Practice, Grand Junction, Colorado; Brighton Salud Family Health Center, Brighton, Colorado; Rose Family Medicine Center, Denver, Colorado; Swedish Family Medicine Center, Littleton, Colorado; AF Williams Family Medicine Center, Denver, Colorado; and La CasaQuigg Newton Health Center, Denver, Colorado. We are also grateful to Elizabeth Staton and Michael Huiras, MD, for their comments on the manuscript. The authors deny any conflict of interest.

Among underserved populations, the burden of tobacco is substantial. There is a clear association between poverty and high rates of tobacco use,^1-3 and smoking is more prevalent among the uninsured (39%) than those with insurance (23%).⁴ Smoking cessation interventions can be successful among low-income and minority patients, especially when tailored to these populations.^5-8 Tobacco counseling, including simple advice to quit, has been shown effective in primary care.^9-11 Since disadvantaged patients, including 63% of the uninsured,¹² are commonly seen in primary care settings, primary care providers are in a unique position to impact tobacco use in underserved patients.

Previous research on cessation counseling rates in low-income patients has yielded conflicting results. Taira and colleagues¹¹ demonstrated that cessation advice by primary care providers was given more frequently to low-income groups. However, this study’s results were based on a written patient questionnaire, and recall may have been a significant limitation. Another study examined physician-reported rates of tobacco cessation counseling, and found that cessation was addressed more frequently with health maintenance organization (HMO)–insured patients (30%) than Medicaid patients (24%).¹³ However, this analysis did not differentiate between primary care providers and specialists, and neither of these studies identified low-income uninsured patients.

It thus remains unclear whether this effective intervention is routinely provided to underserved patients, including the uninsured, in primary care settings. Using a provider survey instrument that clearly identified medically indigent patients, this study examined the frequency with which primary care providers address tobacco use with their Medicaid-insured and uninsured patients compared with those with private or HMO insurance.

Methods

This study was conducted in the 7 primary care practices in the Colorado Research Network (CaReNet) in 1998 and 1999. CaReNet is a state-wide primary care, practice-based research network founded in 1997 with a particular focus on disadvantaged populations, including rural people, minorities, and the urban poor. The practices in CaReNet are affiliated with the University of Colorado Department of Family Medicine. Of the 7 practices, 4 are family medicine residency sites, 2 are federally-funded community health centers, and 1 is a clinic for the medically indigent. The provider mix in CaReNet includes 56% residents (residents average approximately 3 half-day clinics weekly), 21% full-time clinical faculty, 7% private physicians, and 15% other providers (nurse practitioners, physician assistants, and so forth). At the time of our study, none of the practices had a comprehensive tobacco cessation program on site. Colorado Medicaid recipients were eligible for a limited amount of smoking cessation products (this benefit required prior authorization), but Medicaid did not cover comprehensive programs.

A modified version of the 1994 National Ambulatory Medical Care Survey (NAMCS) was administered in each CaReNet practice. The NAMCS instrument is a physician survey that collects information about an ambulatory visit; it has been used by the National Center for Health Statistics since 1973 to analyze trends of ambulatory care. In the context of our study, the key modification was the addition of “uninsured” in the Expected Source of Payment category. This category included patients who were in 1 of several programs that discount charges on the basis of income, thus covering some of the costs of care. All providers received detailed instructions on completing this modified NAMCS form.

Each CaReNet practice collected data on a total of 400 patient visits in 1-week cycles (100 patients per cycle), quarterly, for 1 year. We used the typical NAMCS protocol of collecting data on every second patient presenting for medical care during the study period.¹⁴ The anonymous visit survey forms were coded using standard NAMCS nomenclature. Only patients aged between 13 years and 65 years were included in this analysis because there are almost no uninsured people older than 65 years. To identify patients with private insurance, the options “Private/commercial” and “HMO/other prepaid” were combined (hereafter referred to as “Private/HMO”).

For the present study, we examined the impact of patient insurance on 2 primary outcomes: (1) patient smoking status, and (2) whether smokers received smoking cessation counseling. Each provider coded smoking status as “Yes,” “No,” or “Unknown.” Only patients with a known smoking status (90% of sample) were included in the present analysis. For those patients coded as smokers, we determined whether providers checked the “Smoking Cessation” box.

Analysis

To examine the association between insurance group and study outcomes, we used chi-square tests to determine whether insurance group and other patient demographics (sex, age, ethnicity, and race) were reliably associated with smoking status and cessation counseling. Next, for each primary outcome, we conducted multivariate analyses to examine the effect of patient insurance, while controlling for other important demographic factors (ie, those with P values 0.20 in univariate analyses,¹⁵ as well as additional factors that may account for variability in this relationship. These factors included duration of visit, whether the patient had been seen before in the practice, and whether the patient had at least 1 of the chronic conditions listed on the NAMCS form (hypertension, depression, obesity, or hypercholesterolemia). Because initial random effects analyses revealed no significant practice site effects on the frequency of tobacco use and cessation counseling, all analyses include patient-level data.

The Colorado Multiple Institutional Review Board approved our study design.

Results

Description of sample

CaReNet providers completed NAMCS forms on 2773 patient encounters of 2800 eligible visits (99% completion rate). For this study, of the 2773 encounters, 1443 remained after excluding patients younger than 13 or older than 65 years, and those with sources of payment other than Medicaid, Uninsured, or Private/HMO. As shown in (Table 1), CaReNet patients in the present study were demographically diverse, with a high percentage who were Hispanic (26%), female (74%), or low-income (39% uninsured, 22% Medicaid).

TABLE 1
DEMOGRAPHIC CHARACTERISTICS OF CARENET STUDY SAMPLE

Univariate and multivariate analysis of smoking

A total of 351 patients in the study sample (24%) were identified as smokers. As expected, smoking was significantly more prevalent in the Medicaid and uninsured groups (Table W1*).

(Table 2) presents multivariate logistic regression results showing the significant relationship between insurance and smoking status after controlling for other important demographic and practice variables. Uninsured patients had similar rates of smoking as those with Medicaid; however, smoking among Private/HMO–insured patients was approximately half as frequent as among the uninsured.

In addition to patient insurance, ethnicity and clinical factors predicted whether patients smoked. Non-Hispanic patients were more than twice as likely to be identified as smokers compared with Hispanic patients (P <.001). Also, patients who were new to the practice or who had at least one chronic condition were significantly more likely to be identified as smokers (P = .011 and P = .001, respectively).

Table 2
LOGISTIC REGRESSION RESULTS: RELATIONSHIP OF PATIENT FACTORS WITH LIKELIHOOD OF SMOKING

Univeriate and multivariate analysis of cessation advice or counseling

The second primary analysis examined whether insurance is associated with how often smokers are counseled during visits. Out of 351 smokers, 129 (37%) received tobacco counseling during the medical encounter. Private/HMO insurance and duration of visit were the only factors univariately associated with whether a smoker received counseling (Table W2*).

Multivariate results indicate that patient insurance remained the only significant variable after controlling for other factors that might explain whether smokers received counseling. Smokers with Medicaid were more than twice as likely, and Private/HMO–insured smokers were more than 3 times as likely as uninsured patients (P <.001) to receive smoking cessation counseling (Table 3).

TABLE 3
LOGISTIC REGRESSION RESULTS: PATIENT FACTORS ASSOCIATED WITH LIKELIHOOD OF RECEIVING SMOKING CESSATION COUNSELING

Discussion

These findings demonstrate that although smoking is more common in CaReNet’s Medicaid and uninsured patients, providers gave cessation advice less often to these patients. The actual prevalence of tobacco use may be even greater than we think because providers may underreport it, but our results are similar to national trends.⁴ The decreased rate of tobacco counseling in underserved patients is in contrast to the findings in a study that were based on patient recall,¹¹ rather than the provider-report methodology of NAMCS. However, our counseling results are consistent with a national NAMCS analysis, which found that tobacco use was addressed more frequently with HMO-insured patients than Medicaid patients.¹³ In that study, the overall primary care counseling rate (33%) was similar to that of CaReNet providers (37%). To the best of our knowledge, our finding of a lower rate of tobacco counseling in uninsured patients has not been previously reported.

Our study does not address why providers are less likely to advise Medicaid or uninsured patients to quit smoking. It is possible that tobacco interventions, such as pharmacologic aids and comprehensive cessation programs, may not be available to these groups because of cost. Providers may simply be reflecting this situation by not addressing cessation. Even so, cost and access barriers do not explain why providers would be less likely to give simple cessation advice to disadvantaged smokers. One possibility is that these findings may indicate a lower quality of care for these patients. Other preventive care measures have been shown to be performed less often in uninsured patients,¹⁶ and several studies have documented a lower quality of care for Medicaid and uninsured patients with chronic diseases.^17-19

Limitations

A major limitation of our study is that the uninsured or Medicaid groups may have included sicker or more complex patients at the surveyed visits, thus there may have been less time to devote to tobacco cessation advice during that clinic visit. Unfortunately, the NAMCS instrument does not readily measure disease severity or case mix. In our analysis, we controlled for the presence of 1 or more chronic diseases (limited in NAMCS to 4 specific conditions), but this is only a crude measure of patient complexity. If patients in one of the payment groups were sicker, they might have had more frequent clinic visits, and tobacco cessation may have been addressed at higher rates over time than were found in this cross-sectional study. However, even in the presence of major morbidities, the uninsured often lack continuity because of their tenuous access to care.

If the payer mix of residents and faculty was significantly different, and residents addressed tobacco use at a different rate than faculty, this could explain some of the counseling differences. Unfortunately, this NAMCS instrument is anonymous and cannot identify the type of provider. Similarly, it is possible that the type of visit (acute care, chronic care, or prevention) may account for some of the findings. However, NAMCS also does not specify type of visit and there may be considerable overlap at any given encounter.

Our study administered NAMCS to the practices that make up CaReNet, and the results are not necessarily generalizable to other populations. There is substantial regional variation in health care access programs for the uninsured²⁰; therefore the uninsured patients in CaReNet may not be representative of uninsured in primary care elsewhere. Also, the demographics of CaReNet include higher percentages of Hispanics and Medicaid recipients compared with a national analysis of primary care trends.²¹ CaReNet more closely resembles community health centers,²² except CaReNet has a greater number of Hispanic patients and fewer black patients, reflecting the particular demographics of Colorado. However, the smoking prevalence rates we found in the privately insured, Medicaid, and uninsured groups were similar to national patterns.

Conclusions

Our study argues for the inclusion of a separate payment category that clearly identifies the uninsured in NAMCS and other data collection instruments. Future studies on tobacco counseling rates should be designed to differentiate factors associated with the lower rate of counseling in disadvantaged populations, such as patient complexity, competing demands, lack of access to cessation resources, or lower standards of care. Identification of these factors may be valuable in implementing interventions to improve the rate of counseling for these patients.

If national tobacco goals are to be realized, then socioeconomic disparities in counseling need to be addressed. Our results show that primary care providers can substantially improve the tobacco counseling rate among disadvantaged smokers. As this occurs, the rate of smoking in these patients can be expected to decrease.

*Tables W1 and W2 are available on the JFP Web site, www.jfponline.com.

Acknowledgments

We appreciate the financial support of CaReNet by the University of Colorado School of Medicine Academic Enrichment Fund. We would also like to thank the faculty, residents, and staff at the following CaReNet sites for their assistance with this study: CU Care, Denver, Colorado; St. Mary’s Family Practice, Grand Junction, Colorado; Brighton Salud Family Health Center, Brighton, Colorado; Rose Family Medicine Center, Denver, Colorado; Swedish Family Medicine Center, Littleton, Colorado; AF Williams Family Medicine Center, Denver, Colorado; and La CasaQuigg Newton Health Center, Denver, Colorado. We are also grateful to Elizabeth Staton and Michael Huiras, MD, for their comments on the manuscript. The authors deny any conflict of interest.

Acute otitis media (AOM) is the most frequent diagnosis for which antibiotics are prescribed in the United States^1,2 and is among the most frequent in France.³ The benefit of antibiotic treatment is under scrutiny as physicians, parents, and policymakers in the United States,^4-7 France,^3,8-13 and elsewhere^14-16 become increasingly worried about bacterial resistance to antibiotics.

Patients and caregivers often have a firm idea of what is wrong with their child and how it should be treated.^17,18 Conflict with the physician may partially result from different interpretations that parents and physicians give to the same cues from the child’s history and physical examination, and the different implications derived from them. Despite evidence to the contrary,^19-22 studies in the United States and Britain show that physicians feel parents are becoming increasingly demanding: They expect them to prescribe antibiotics even when they are not indicated and will be dissatisfied if they are not prescribed.^22-29

The purpose of our study was to elucidate the differences between physicians and parents both in the United States and France by comparing their diagnostic judgments and treatment choices when dealing with children who might be suffering from AOM.

Methods

This study was primarily a “judgment analysis.”³⁰ In judgment analyses, individuals make a series of judgments (eg, diagnoses or treatment choices about a series of patients) according to a set of varying cues (eg, signs and symptoms). Using multiple linear regression analyses, it is possible to determine how much weight they put on each cue when making their decisions. Separate regression analyses are performed for each person, with the cues as the predictors and the judgment (diagnoses or treatment decisions) as the variable predicted. The beta weights from the regression, suitably adjusted for differences in units of measurement, are estimates of the relative importance of the cues in making judgments. A higher beta weight for a cue indicates that it carried greater weight in making the diagnosis or treatment decision. Between- and within-group comparisons of these weights may reveal why people differ in their judgments of identical cases.

Participants

The samples consisted of 19 US family physicians, 35 French general practitioners, 21 French pediatricians, 52 US parents, and 86 French parents. The US physicians were recruited from the 62 family physicians in the region of Albany and Schenectady, New York, through written and oral appeals from the study team. The French practitioners consisted of personal contacts and members of 2 networks of research minded general practitioners. The French pediatricians were recruited by written appeals to the 30 practicing pediatricians in and around Tours, France. The US parents were responders to a recruitment letter sent to 100 randomly selected parents of young children of one primary care practice in suburban Albany. The French parents were recruited by students at the Université François-Rabelais in Tours.³¹

Procedure

We constructed 46 paper scenarios describing hypothetical children aged 15 months who might have ear infections. Each scenario displayed the values for 15 different cues, such as fever, redness of the tympanic membrane, or ear pain during examination (see the figure at www.jfponline.com for more information). These cues were selected after consultation with physicians and parents, who felt that they were important for decisions about the diagnosis and treatment of AOM. The diagnostic cues were based on our previous study of US pediatricians³² and on the medical literature in the United States,^33-35 France,^36,37 and elsewhere.^14,38 We included the result of insufflation of the tympanic membrane, even though French primary care physicians are not taught to use pneumatic insufflation in their diagnosis of AOM.

The cue values for each case were generated randomly using a computer program. Very implausible combinations of cue values were excluded from the scenarios. We did not, however, attempt to create a set of cases that would reproduce the actual mix of cases seen by pediatricians in their offices. Indeed, the inclusion of unlikely combinations of cue values was useful in forcing the participants to choose which cues were most important. Intercorrelation between cues was small to moderate, ranging from -0.40 to 0.39. This resulted from the rules for excluding implausible cases (eg, 0.39 between bulging and mobility) or from chance (eg, -0.40 between a history of ear pain and ear pain during the examination). The participants were told to put themselves in the place of the examining physician. They were presented twice with the same 46 scenarios. For one set they were asked to judge the probability that each child had AOM. For the other, in which the cases were presented in a different order, they were to decide whether to treat with antibiotics or to observe the child. They were also asked in the second set to indicate on a 5-point scale their degree of certainty that this was the right choice. Half the participants completed the diagnosis set first; half did the treatment set first. They completed the study at home. They were instructed to refrain from looking back at the first set after finishing and to take a short break of up to a day between the 2 sets. At the end of the study session, the participants answered certain questions about attitudes toward health care and risk and about their background that might account for differences in their responses to the scenarios.*

Data analysis

Two multiple linear regression analyses were performed for each participant (physician or parent). In one regression, the predicted variable was the judgment of the probability of AOM; in the other, it was the choice of treatment. Before performing the second regression, the treatment choice was combined with the degree of certainty about it to create a treatment score on a 10-point scale ranging from -5 for “observe/completely sure” to 5 for “treat/completely sure” (with no 0). This score was used in the multiple linear regression analysis as the dependent variable. A participant was included in these analyses only if his or her R² passed the F-test for fitness of the multiple regression model at P less than .05; failure of a model to pass the F-test meant that the individual’s judgments were not predictable by the cues. This could occur if he or she had answered randomly or had made mistakes.

Multivariate analysis of variance was used to test the differences in the mean responses to the questions on attitudes and opinions of each group of participants. The association of individual attitudes and opinions about AOM and health care with treatment choices was explored by correlating the percentage of cases treated with antibiotics with the responses to each of the questions about attitudes and opinions.

Results

The judgments of the parents were remarkably similar to those of the physicians, both in the United States and France. The means and the ranges of the mean probability judgments of the individual participants in all the groups were almost identical at 50% or just above (Table 1, row 1). The cue weights for diagnosis (Table 2) were also quite similar. The physical examination provided the key information for both the parents and physicians. The only differences were that French parents gave more weight than the other groups to a past history of ear infections (beta weight = 0.15) and focused more on fever (0.30) and ear pain during the examination (0.28) than on bulging (0.17).

In choosing treatment, the parents in each country had lower mean treatment scores (Table 1, row 2) than the physicians, but these differences were not statistically significant. Likewise, the overall percentages of cases judged as needing antibiotics, though lower for parents, were not significantly different between the physicians and parents of each country: 53.0% for US physicians, 44.6% for US parents, 53.4% for French physicians, and 51.1% for French parents. Among the 4 groups the important cues for treatment (Table 3) were similar, stressing the physical examination findings and de-emphasizing the history, including the parents’ report of ear pain. French parents gave atypical stress to the child’s temperature (beta weight = 0.52), so that they differed from other groups in their judgments of several individual scenarios. Both groups of parents assigned significantly less weight than the physicians of all groups to the cue of the parents’ position on antibiotics. Indeed, when asked to put themselves in the place of a physician, the parents did not give importance to any of the parent-sensitive cues, even when the question was treatment. These cues included ear pain noted by the parent, the parents’ personal position concerning antibiotics, whether caring for a sick child greatly upsets the parents’ ordinary schedule, and whether there are babies or other small children in the family.

Attitudes and opinions related to AOM and health care were not associated with the percentage of cases in which the participant of any group chose to treat with antibiotics. Among both the Americans and the French, the parents thought physicians worried more about charges of malpractice when making decisions about patient care than physicians claimed they did (P <.005 for both countries). Parents were bothered when their child was sick more than physicians were bothered when “one of my patients” was sick (P <.005 for both). If there was a possibility that their child had “a serious illness that is rare but curable,” parents were more willing than physicians to order diagnostic tests even when they would cost the parents “a great deal of time and/or money” (P <.005 for both). Interestingly, parents agreed less strongly than physicians with the statement “A doctor should pay close attention to the needs and preferences of a child’s parents” (P <.005 for both). US parents, but not those in France, agreed more strongly than physicians that if their child might have a serious illness that was rare but curable diagnostic tests should be ordered even when they were “very expensive for the child’s insurance plan” (P <.005). French parents, but not American ones, agreed more strongly than physicians that “all ear infections should be treated with antibiotics” (P <.005).

TABLE 1
DIFFERENCES AMONG GROUPS OF PARTICIPANTS

TABLE 2
MEAN BETA WEIGHTS FOR CUES FOR JUDGING PROBABILITY OF ACUTE OTITIS MEDIA

TABLE 3
MEAN BETA WEIGHTS FOR CUES FOR CHOOSING TREATMENT OF ACUTE OTITIS MEDIA

Discussion

Although physicians are aware that antibiotic resistance of bacteria is an increasing problem,^39-41 they continue to prescribe antibiotics for patients who are unlikely to benefit from them.^{2,3,13,39,42,43} There are multiple plausible reasons for this.^7,17,44-46 Some of these relate to physicians’ perceptions of the wants and needs of their patients and their caretakers. Physicians may⁴⁷ (or may not⁴⁸) make different decisions for individuals they are dealing with than for community groups. They know the public misunderstands the indications for antibiotics,^25,49,50 and they may perceive, often incorrectly,^19,21 that patients or parents want antibiotics and will be dissatisfied if they do not receive them.^{22,23,26,29,45,51} They may practice defensive medicine²⁸ or believe that it takes less time and effort to prescribe antibiotics than to explain why they are withholding them.^28,45 They may be sensitive to the socioeconomic pressures on patients and parents related to daycare policies, work schedules, and the costs of return visits.⁴⁶

Our study results should be reassuring to physicians. The most striking finding was the similarity between parents’ diagnostic judgments and treatment choices and those of physicians. The only difference in judgment policies was that the French parents placed greater stress than the French physicians on fever and gave a lower weight to bulging when deciding about treatment, which may be understandable given their unfamiliarity with the technical aspects of examining an ear.³¹ Indeed, contrary to our expectations, parents in both countries gave less weight to parent-sensitive cues than did physicians. Overall, parents did not diagnose AOM in the scenario children or recommend treating them with antibiotics more frequently than physicians.

The parents’ restraint concerning antibiotics is surprising given physicians’ perceptions and the results of a recent US survey⁵² in which 96% of parents said that “ear infections can affect a baby’s hearing,” and only 11% thought that “most ear infections get better by themselves.” Also, recent experts have stressed the high indirect costs of an episode of AOM and the value to parents of reducing the duration of the illness.^53-56 One explanation may be the rising parental worry that antibiotic treatment may lessen their child’s ability to fight off future infections, in particular because of the spread of antibiotic-resistant bacteria.²⁷ The parents in our study worried as much as the physicians about the adverse effects of antibiotics and agreed just as strongly as the physicians that the resistance of bacteria to antibiotics is the most important threat to the future health of the public. The parents’ ability to adopt the physician’s point of view should encourage physicians to undertake the patient and parent education efforts recently recommended^25,28,57 as the best way to reduce the excessive prescription of antibiotics.

We had anticipated incorrectly that the decision to treat would be influenced by attitudes toward uncertainty, ambiguity, and risk.^59,63 We had also expected, again incorrectly, that a greater belief in the usefulness of antibiotics and (for the parents alone) in the contagious nature of ear infections — and a lesser worry about antibiotic side effects and bacterial resistance — would identify physicians and parents who opted more frequently for antibiotics. The explanation may be that our questions were insensitive or that general attitudes are poor predictors of individual case-by-case choices and behavior. It may also be that physicians—and even parents who take the role of physicians—believe that diagnosis is the first and determining step in managing a possible ear infection.

Limitations

Our study has several limitations. First, generalization is limited by the small samples, the inequality of the sample sizes, and the convenience nature of the samples. Second, the patients were hypothetical, presented on paper in schematic form, with neither the richness nor the vividness of the real children brought by parents to physicians’ offices. Although the use of “paper patients” has been questioned,^64,65 it is practical and has been supported in other studies of clinical decision making.^66-69 Third, comparisons between the French and Americans may have been influenced by unappreciated differences in meaning of the French and English versions of the scenarios and questions.

Conclusions

It is encouraging that parents in our study were able to adopt the physician’s perspective and to focus on medical indications rather than on parental needs in their treatment decisions, that they did not choose to prescribe antibiotics more frequently than the physicians, and that they were as concerned as the physicians about the adverse effects of antibiotics and the threat from resistant bacteria. Patients and parents may, therefore, be more willing to forgo antibiotics than physicians realize.

Acknowledgments

We thank the following for their invaluable advice and assistance: Bernard Grenier, MD; Héléne Touchon, MD; Joél Cogneau, MD; Marie-Ange Lecomte, MD; Sabine Maciaszczyk-Jedeau, PhD; Appleton Mason, MD; the Association des Pédiatres de Ville en Touraine; the Latham Medical Group; the Albany Medical Center Department of Family Practice; Héma Sandanam; and Roberta Sandler, RN.

Acute otitis media (AOM) is the most frequent diagnosis for which antibiotics are prescribed in the United States^1,2 and is among the most frequent in France.³ The benefit of antibiotic treatment is under scrutiny as physicians, parents, and policymakers in the United States,^4-7 France,^3,8-13 and elsewhere^14-16 become increasingly worried about bacterial resistance to antibiotics.

Patients and caregivers often have a firm idea of what is wrong with their child and how it should be treated.^17,18 Conflict with the physician may partially result from different interpretations that parents and physicians give to the same cues from the child’s history and physical examination, and the different implications derived from them. Despite evidence to the contrary,^19-22 studies in the United States and Britain show that physicians feel parents are becoming increasingly demanding: They expect them to prescribe antibiotics even when they are not indicated and will be dissatisfied if they are not prescribed.^22-29

The purpose of our study was to elucidate the differences between physicians and parents both in the United States and France by comparing their diagnostic judgments and treatment choices when dealing with children who might be suffering from AOM.

Methods

This study was primarily a “judgment analysis.”³⁰ In judgment analyses, individuals make a series of judgments (eg, diagnoses or treatment choices about a series of patients) according to a set of varying cues (eg, signs and symptoms). Using multiple linear regression analyses, it is possible to determine how much weight they put on each cue when making their decisions. Separate regression analyses are performed for each person, with the cues as the predictors and the judgment (diagnoses or treatment decisions) as the variable predicted. The beta weights from the regression, suitably adjusted for differences in units of measurement, are estimates of the relative importance of the cues in making judgments. A higher beta weight for a cue indicates that it carried greater weight in making the diagnosis or treatment decision. Between- and within-group comparisons of these weights may reveal why people differ in their judgments of identical cases.

Participants

The samples consisted of 19 US family physicians, 35 French general practitioners, 21 French pediatricians, 52 US parents, and 86 French parents. The US physicians were recruited from the 62 family physicians in the region of Albany and Schenectady, New York, through written and oral appeals from the study team. The French practitioners consisted of personal contacts and members of 2 networks of research minded general practitioners. The French pediatricians were recruited by written appeals to the 30 practicing pediatricians in and around Tours, France. The US parents were responders to a recruitment letter sent to 100 randomly selected parents of young children of one primary care practice in suburban Albany. The French parents were recruited by students at the Université François-Rabelais in Tours.³¹

Procedure

We constructed 46 paper scenarios describing hypothetical children aged 15 months who might have ear infections. Each scenario displayed the values for 15 different cues, such as fever, redness of the tympanic membrane, or ear pain during examination (see the figure at www.jfponline.com for more information). These cues were selected after consultation with physicians and parents, who felt that they were important for decisions about the diagnosis and treatment of AOM. The diagnostic cues were based on our previous study of US pediatricians³² and on the medical literature in the United States,^33-35 France,^36,37 and elsewhere.^14,38 We included the result of insufflation of the tympanic membrane, even though French primary care physicians are not taught to use pneumatic insufflation in their diagnosis of AOM.

The cue values for each case were generated randomly using a computer program. Very implausible combinations of cue values were excluded from the scenarios. We did not, however, attempt to create a set of cases that would reproduce the actual mix of cases seen by pediatricians in their offices. Indeed, the inclusion of unlikely combinations of cue values was useful in forcing the participants to choose which cues were most important. Intercorrelation between cues was small to moderate, ranging from -0.40 to 0.39. This resulted from the rules for excluding implausible cases (eg, 0.39 between bulging and mobility) or from chance (eg, -0.40 between a history of ear pain and ear pain during the examination). The participants were told to put themselves in the place of the examining physician. They were presented twice with the same 46 scenarios. For one set they were asked to judge the probability that each child had AOM. For the other, in which the cases were presented in a different order, they were to decide whether to treat with antibiotics or to observe the child. They were also asked in the second set to indicate on a 5-point scale their degree of certainty that this was the right choice. Half the participants completed the diagnosis set first; half did the treatment set first. They completed the study at home. They were instructed to refrain from looking back at the first set after finishing and to take a short break of up to a day between the 2 sets. At the end of the study session, the participants answered certain questions about attitudes toward health care and risk and about their background that might account for differences in their responses to the scenarios.*

Data analysis

Two multiple linear regression analyses were performed for each participant (physician or parent). In one regression, the predicted variable was the judgment of the probability of AOM; in the other, it was the choice of treatment. Before performing the second regression, the treatment choice was combined with the degree of certainty about it to create a treatment score on a 10-point scale ranging from -5 for “observe/completely sure” to 5 for “treat/completely sure” (with no 0). This score was used in the multiple linear regression analysis as the dependent variable. A participant was included in these analyses only if his or her R² passed the F-test for fitness of the multiple regression model at P less than .05; failure of a model to pass the F-test meant that the individual’s judgments were not predictable by the cues. This could occur if he or she had answered randomly or had made mistakes.

Multivariate analysis of variance was used to test the differences in the mean responses to the questions on attitudes and opinions of each group of participants. The association of individual attitudes and opinions about AOM and health care with treatment choices was explored by correlating the percentage of cases treated with antibiotics with the responses to each of the questions about attitudes and opinions.

Results

The judgments of the parents were remarkably similar to those of the physicians, both in the United States and France. The means and the ranges of the mean probability judgments of the individual participants in all the groups were almost identical at 50% or just above (Table 1, row 1). The cue weights for diagnosis (Table 2) were also quite similar. The physical examination provided the key information for both the parents and physicians. The only differences were that French parents gave more weight than the other groups to a past history of ear infections (beta weight = 0.15) and focused more on fever (0.30) and ear pain during the examination (0.28) than on bulging (0.17).

In choosing treatment, the parents in each country had lower mean treatment scores (Table 1, row 2) than the physicians, but these differences were not statistically significant. Likewise, the overall percentages of cases judged as needing antibiotics, though lower for parents, were not significantly different between the physicians and parents of each country: 53.0% for US physicians, 44.6% for US parents, 53.4% for French physicians, and 51.1% for French parents. Among the 4 groups the important cues for treatment (Table 3) were similar, stressing the physical examination findings and de-emphasizing the history, including the parents’ report of ear pain. French parents gave atypical stress to the child’s temperature (beta weight = 0.52), so that they differed from other groups in their judgments of several individual scenarios. Both groups of parents assigned significantly less weight than the physicians of all groups to the cue of the parents’ position on antibiotics. Indeed, when asked to put themselves in the place of a physician, the parents did not give importance to any of the parent-sensitive cues, even when the question was treatment. These cues included ear pain noted by the parent, the parents’ personal position concerning antibiotics, whether caring for a sick child greatly upsets the parents’ ordinary schedule, and whether there are babies or other small children in the family.

Attitudes and opinions related to AOM and health care were not associated with the percentage of cases in which the participant of any group chose to treat with antibiotics. Among both the Americans and the French, the parents thought physicians worried more about charges of malpractice when making decisions about patient care than physicians claimed they did (P <.005 for both countries). Parents were bothered when their child was sick more than physicians were bothered when “one of my patients” was sick (P <.005 for both). If there was a possibility that their child had “a serious illness that is rare but curable,” parents were more willing than physicians to order diagnostic tests even when they would cost the parents “a great deal of time and/or money” (P <.005 for both). Interestingly, parents agreed less strongly than physicians with the statement “A doctor should pay close attention to the needs and preferences of a child’s parents” (P <.005 for both). US parents, but not those in France, agreed more strongly than physicians that if their child might have a serious illness that was rare but curable diagnostic tests should be ordered even when they were “very expensive for the child’s insurance plan” (P <.005). French parents, but not American ones, agreed more strongly than physicians that “all ear infections should be treated with antibiotics” (P <.005).

TABLE 1
DIFFERENCES AMONG GROUPS OF PARTICIPANTS

TABLE 2
MEAN BETA WEIGHTS FOR CUES FOR JUDGING PROBABILITY OF ACUTE OTITIS MEDIA

TABLE 3
MEAN BETA WEIGHTS FOR CUES FOR CHOOSING TREATMENT OF ACUTE OTITIS MEDIA

Discussion

Although physicians are aware that antibiotic resistance of bacteria is an increasing problem,^39-41 they continue to prescribe antibiotics for patients who are unlikely to benefit from them.^{2,3,13,39,42,43} There are multiple plausible reasons for this.^7,17,44-46 Some of these relate to physicians’ perceptions of the wants and needs of their patients and their caretakers. Physicians may⁴⁷ (or may not⁴⁸) make different decisions for individuals they are dealing with than for community groups. They know the public misunderstands the indications for antibiotics,^25,49,50 and they may perceive, often incorrectly,^19,21 that patients or parents want antibiotics and will be dissatisfied if they do not receive them.^{22,23,26,29,45,51} They may practice defensive medicine²⁸ or believe that it takes less time and effort to prescribe antibiotics than to explain why they are withholding them.^28,45 They may be sensitive to the socioeconomic pressures on patients and parents related to daycare policies, work schedules, and the costs of return visits.⁴⁶

Our study results should be reassuring to physicians. The most striking finding was the similarity between parents’ diagnostic judgments and treatment choices and those of physicians. The only difference in judgment policies was that the French parents placed greater stress than the French physicians on fever and gave a lower weight to bulging when deciding about treatment, which may be understandable given their unfamiliarity with the technical aspects of examining an ear.³¹ Indeed, contrary to our expectations, parents in both countries gave less weight to parent-sensitive cues than did physicians. Overall, parents did not diagnose AOM in the scenario children or recommend treating them with antibiotics more frequently than physicians.

The parents’ restraint concerning antibiotics is surprising given physicians’ perceptions and the results of a recent US survey⁵² in which 96% of parents said that “ear infections can affect a baby’s hearing,” and only 11% thought that “most ear infections get better by themselves.” Also, recent experts have stressed the high indirect costs of an episode of AOM and the value to parents of reducing the duration of the illness.^53-56 One explanation may be the rising parental worry that antibiotic treatment may lessen their child’s ability to fight off future infections, in particular because of the spread of antibiotic-resistant bacteria.²⁷ The parents in our study worried as much as the physicians about the adverse effects of antibiotics and agreed just as strongly as the physicians that the resistance of bacteria to antibiotics is the most important threat to the future health of the public. The parents’ ability to adopt the physician’s point of view should encourage physicians to undertake the patient and parent education efforts recently recommended^25,28,57 as the best way to reduce the excessive prescription of antibiotics.

We had anticipated incorrectly that the decision to treat would be influenced by attitudes toward uncertainty, ambiguity, and risk.^59,63 We had also expected, again incorrectly, that a greater belief in the usefulness of antibiotics and (for the parents alone) in the contagious nature of ear infections — and a lesser worry about antibiotic side effects and bacterial resistance — would identify physicians and parents who opted more frequently for antibiotics. The explanation may be that our questions were insensitive or that general attitudes are poor predictors of individual case-by-case choices and behavior. It may also be that physicians—and even parents who take the role of physicians—believe that diagnosis is the first and determining step in managing a possible ear infection.

Limitations

Our study has several limitations. First, generalization is limited by the small samples, the inequality of the sample sizes, and the convenience nature of the samples. Second, the patients were hypothetical, presented on paper in schematic form, with neither the richness nor the vividness of the real children brought by parents to physicians’ offices. Although the use of “paper patients” has been questioned,^64,65 it is practical and has been supported in other studies of clinical decision making.^66-69 Third, comparisons between the French and Americans may have been influenced by unappreciated differences in meaning of the French and English versions of the scenarios and questions.

Conclusions

It is encouraging that parents in our study were able to adopt the physician’s perspective and to focus on medical indications rather than on parental needs in their treatment decisions, that they did not choose to prescribe antibiotics more frequently than the physicians, and that they were as concerned as the physicians about the adverse effects of antibiotics and the threat from resistant bacteria. Patients and parents may, therefore, be more willing to forgo antibiotics than physicians realize.

Acknowledgments

We thank the following for their invaluable advice and assistance: Bernard Grenier, MD; Héléne Touchon, MD; Joél Cogneau, MD; Marie-Ange Lecomte, MD; Sabine Maciaszczyk-Jedeau, PhD; Appleton Mason, MD; the Association des Pédiatres de Ville en Touraine; the Latham Medical Group; the Albany Medical Center Department of Family Practice; Héma Sandanam; and Roberta Sandler, RN.

Acute otitis media (AOM) is the most frequent diagnosis for which antibiotics are prescribed in the United States^1,2 and is among the most frequent in France.³ The benefit of antibiotic treatment is under scrutiny as physicians, parents, and policymakers in the United States,^4-7 France,^3,8-13 and elsewhere^14-16 become increasingly worried about bacterial resistance to antibiotics.

Patients and caregivers often have a firm idea of what is wrong with their child and how it should be treated.^17,18 Conflict with the physician may partially result from different interpretations that parents and physicians give to the same cues from the child’s history and physical examination, and the different implications derived from them. Despite evidence to the contrary,^19-22 studies in the United States and Britain show that physicians feel parents are becoming increasingly demanding: They expect them to prescribe antibiotics even when they are not indicated and will be dissatisfied if they are not prescribed.^22-29

The purpose of our study was to elucidate the differences between physicians and parents both in the United States and France by comparing their diagnostic judgments and treatment choices when dealing with children who might be suffering from AOM.

Methods

This study was primarily a “judgment analysis.”³⁰ In judgment analyses, individuals make a series of judgments (eg, diagnoses or treatment choices about a series of patients) according to a set of varying cues (eg, signs and symptoms). Using multiple linear regression analyses, it is possible to determine how much weight they put on each cue when making their decisions. Separate regression analyses are performed for each person, with the cues as the predictors and the judgment (diagnoses or treatment decisions) as the variable predicted. The beta weights from the regression, suitably adjusted for differences in units of measurement, are estimates of the relative importance of the cues in making judgments. A higher beta weight for a cue indicates that it carried greater weight in making the diagnosis or treatment decision. Between- and within-group comparisons of these weights may reveal why people differ in their judgments of identical cases.

Participants

The samples consisted of 19 US family physicians, 35 French general practitioners, 21 French pediatricians, 52 US parents, and 86 French parents. The US physicians were recruited from the 62 family physicians in the region of Albany and Schenectady, New York, through written and oral appeals from the study team. The French practitioners consisted of personal contacts and members of 2 networks of research minded general practitioners. The French pediatricians were recruited by written appeals to the 30 practicing pediatricians in and around Tours, France. The US parents were responders to a recruitment letter sent to 100 randomly selected parents of young children of one primary care practice in suburban Albany. The French parents were recruited by students at the Université François-Rabelais in Tours.³¹

Procedure

We constructed 46 paper scenarios describing hypothetical children aged 15 months who might have ear infections. Each scenario displayed the values for 15 different cues, such as fever, redness of the tympanic membrane, or ear pain during examination (see the figure at www.jfponline.com for more information). These cues were selected after consultation with physicians and parents, who felt that they were important for decisions about the diagnosis and treatment of AOM. The diagnostic cues were based on our previous study of US pediatricians³² and on the medical literature in the United States,^33-35 France,^36,37 and elsewhere.^14,38 We included the result of insufflation of the tympanic membrane, even though French primary care physicians are not taught to use pneumatic insufflation in their diagnosis of AOM.

The cue values for each case were generated randomly using a computer program. Very implausible combinations of cue values were excluded from the scenarios. We did not, however, attempt to create a set of cases that would reproduce the actual mix of cases seen by pediatricians in their offices. Indeed, the inclusion of unlikely combinations of cue values was useful in forcing the participants to choose which cues were most important. Intercorrelation between cues was small to moderate, ranging from -0.40 to 0.39. This resulted from the rules for excluding implausible cases (eg, 0.39 between bulging and mobility) or from chance (eg, -0.40 between a history of ear pain and ear pain during the examination). The participants were told to put themselves in the place of the examining physician. They were presented twice with the same 46 scenarios. For one set they were asked to judge the probability that each child had AOM. For the other, in which the cases were presented in a different order, they were to decide whether to treat with antibiotics or to observe the child. They were also asked in the second set to indicate on a 5-point scale their degree of certainty that this was the right choice. Half the participants completed the diagnosis set first; half did the treatment set first. They completed the study at home. They were instructed to refrain from looking back at the first set after finishing and to take a short break of up to a day between the 2 sets. At the end of the study session, the participants answered certain questions about attitudes toward health care and risk and about their background that might account for differences in their responses to the scenarios.*

Data analysis

Two multiple linear regression analyses were performed for each participant (physician or parent). In one regression, the predicted variable was the judgment of the probability of AOM; in the other, it was the choice of treatment. Before performing the second regression, the treatment choice was combined with the degree of certainty about it to create a treatment score on a 10-point scale ranging from -5 for “observe/completely sure” to 5 for “treat/completely sure” (with no 0). This score was used in the multiple linear regression analysis as the dependent variable. A participant was included in these analyses only if his or her R² passed the F-test for fitness of the multiple regression model at P less than .05; failure of a model to pass the F-test meant that the individual’s judgments were not predictable by the cues. This could occur if he or she had answered randomly or had made mistakes.

Multivariate analysis of variance was used to test the differences in the mean responses to the questions on attitudes and opinions of each group of participants. The association of individual attitudes and opinions about AOM and health care with treatment choices was explored by correlating the percentage of cases treated with antibiotics with the responses to each of the questions about attitudes and opinions.

Results

The judgments of the parents were remarkably similar to those of the physicians, both in the United States and France. The means and the ranges of the mean probability judgments of the individual participants in all the groups were almost identical at 50% or just above (Table 1, row 1). The cue weights for diagnosis (Table 2) were also quite similar. The physical examination provided the key information for both the parents and physicians. The only differences were that French parents gave more weight than the other groups to a past history of ear infections (beta weight = 0.15) and focused more on fever (0.30) and ear pain during the examination (0.28) than on bulging (0.17).

In choosing treatment, the parents in each country had lower mean treatment scores (Table 1, row 2) than the physicians, but these differences were not statistically significant. Likewise, the overall percentages of cases judged as needing antibiotics, though lower for parents, were not significantly different between the physicians and parents of each country: 53.0% for US physicians, 44.6% for US parents, 53.4% for French physicians, and 51.1% for French parents. Among the 4 groups the important cues for treatment (Table 3) were similar, stressing the physical examination findings and de-emphasizing the history, including the parents’ report of ear pain. French parents gave atypical stress to the child’s temperature (beta weight = 0.52), so that they differed from other groups in their judgments of several individual scenarios. Both groups of parents assigned significantly less weight than the physicians of all groups to the cue of the parents’ position on antibiotics. Indeed, when asked to put themselves in the place of a physician, the parents did not give importance to any of the parent-sensitive cues, even when the question was treatment. These cues included ear pain noted by the parent, the parents’ personal position concerning antibiotics, whether caring for a sick child greatly upsets the parents’ ordinary schedule, and whether there are babies or other small children in the family.

Attitudes and opinions related to AOM and health care were not associated with the percentage of cases in which the participant of any group chose to treat with antibiotics. Among both the Americans and the French, the parents thought physicians worried more about charges of malpractice when making decisions about patient care than physicians claimed they did (P <.005 for both countries). Parents were bothered when their child was sick more than physicians were bothered when “one of my patients” was sick (P <.005 for both). If there was a possibility that their child had “a serious illness that is rare but curable,” parents were more willing than physicians to order diagnostic tests even when they would cost the parents “a great deal of time and/or money” (P <.005 for both). Interestingly, parents agreed less strongly than physicians with the statement “A doctor should pay close attention to the needs and preferences of a child’s parents” (P <.005 for both). US parents, but not those in France, agreed more strongly than physicians that if their child might have a serious illness that was rare but curable diagnostic tests should be ordered even when they were “very expensive for the child’s insurance plan” (P <.005). French parents, but not American ones, agreed more strongly than physicians that “all ear infections should be treated with antibiotics” (P <.005).

TABLE 1
DIFFERENCES AMONG GROUPS OF PARTICIPANTS

TABLE 2
MEAN BETA WEIGHTS FOR CUES FOR JUDGING PROBABILITY OF ACUTE OTITIS MEDIA

TABLE 3
MEAN BETA WEIGHTS FOR CUES FOR CHOOSING TREATMENT OF ACUTE OTITIS MEDIA

Discussion

Although physicians are aware that antibiotic resistance of bacteria is an increasing problem,^39-41 they continue to prescribe antibiotics for patients who are unlikely to benefit from them.^{2,3,13,39,42,43} There are multiple plausible reasons for this.^7,17,44-46 Some of these relate to physicians’ perceptions of the wants and needs of their patients and their caretakers. Physicians may⁴⁷ (or may not⁴⁸) make different decisions for individuals they are dealing with than for community groups. They know the public misunderstands the indications for antibiotics,^25,49,50 and they may perceive, often incorrectly,^19,21 that patients or parents want antibiotics and will be dissatisfied if they do not receive them.^{22,23,26,29,45,51} They may practice defensive medicine²⁸ or believe that it takes less time and effort to prescribe antibiotics than to explain why they are withholding them.^28,45 They may be sensitive to the socioeconomic pressures on patients and parents related to daycare policies, work schedules, and the costs of return visits.⁴⁶

Our study results should be reassuring to physicians. The most striking finding was the similarity between parents’ diagnostic judgments and treatment choices and those of physicians. The only difference in judgment policies was that the French parents placed greater stress than the French physicians on fever and gave a lower weight to bulging when deciding about treatment, which may be understandable given their unfamiliarity with the technical aspects of examining an ear.³¹ Indeed, contrary to our expectations, parents in both countries gave less weight to parent-sensitive cues than did physicians. Overall, parents did not diagnose AOM in the scenario children or recommend treating them with antibiotics more frequently than physicians.

The parents’ restraint concerning antibiotics is surprising given physicians’ perceptions and the results of a recent US survey⁵² in which 96% of parents said that “ear infections can affect a baby’s hearing,” and only 11% thought that “most ear infections get better by themselves.” Also, recent experts have stressed the high indirect costs of an episode of AOM and the value to parents of reducing the duration of the illness.^53-56 One explanation may be the rising parental worry that antibiotic treatment may lessen their child’s ability to fight off future infections, in particular because of the spread of antibiotic-resistant bacteria.²⁷ The parents in our study worried as much as the physicians about the adverse effects of antibiotics and agreed just as strongly as the physicians that the resistance of bacteria to antibiotics is the most important threat to the future health of the public. The parents’ ability to adopt the physician’s point of view should encourage physicians to undertake the patient and parent education efforts recently recommended^25,28,57 as the best way to reduce the excessive prescription of antibiotics.

We had anticipated incorrectly that the decision to treat would be influenced by attitudes toward uncertainty, ambiguity, and risk.^59,63 We had also expected, again incorrectly, that a greater belief in the usefulness of antibiotics and (for the parents alone) in the contagious nature of ear infections — and a lesser worry about antibiotic side effects and bacterial resistance — would identify physicians and parents who opted more frequently for antibiotics. The explanation may be that our questions were insensitive or that general attitudes are poor predictors of individual case-by-case choices and behavior. It may also be that physicians—and even parents who take the role of physicians—believe that diagnosis is the first and determining step in managing a possible ear infection.

Limitations

Our study has several limitations. First, generalization is limited by the small samples, the inequality of the sample sizes, and the convenience nature of the samples. Second, the patients were hypothetical, presented on paper in schematic form, with neither the richness nor the vividness of the real children brought by parents to physicians’ offices. Although the use of “paper patients” has been questioned,^64,65 it is practical and has been supported in other studies of clinical decision making.^66-69 Third, comparisons between the French and Americans may have been influenced by unappreciated differences in meaning of the French and English versions of the scenarios and questions.

Conclusions

It is encouraging that parents in our study were able to adopt the physician’s perspective and to focus on medical indications rather than on parental needs in their treatment decisions, that they did not choose to prescribe antibiotics more frequently than the physicians, and that they were as concerned as the physicians about the adverse effects of antibiotics and the threat from resistant bacteria. Patients and parents may, therefore, be more willing to forgo antibiotics than physicians realize.

Acknowledgments

We thank the following for their invaluable advice and assistance: Bernard Grenier, MD; Héléne Touchon, MD; Joél Cogneau, MD; Marie-Ange Lecomte, MD; Sabine Maciaszczyk-Jedeau, PhD; Appleton Mason, MD; the Association des Pédiatres de Ville en Touraine; the Latham Medical Group; the Albany Medical Center Department of Family Practice; Héma Sandanam; and Roberta Sandler, RN.