Item Development for the Physician‐PREPARED

Australian investigators designed the Physician‐PREPARED survey instrument to measure the quality of discharge planning activities and communication. The investigators developed the survey with the following process that was not published previously. First, a literature review identified survey content germane to outpatient practitioners.10 Investigators conducted interviews, focus groups, and pilot surveys to prioritize items for the survey instrument. The volunteer subjects for item development were general medical practitioners in Adelaide and Sydney, the capital cities of two states in Australia. The draft instrument was circulated to a small group of general medical practitioners for comment on layout, wording, and question intent. After feedback, minor modifications were made to item content and response categories. The result of development in Australia was a survey instrument with 15 items (see Appendix). The items reflected the following key areas of discharge quality: timeliness of communication, patient health status at discharge, adequacy of discharge support services, discharge medication information, and reasons for medication changes. These areas were congruent with the results of other investigators who assessed the quality of discharge planning and communication.14, 15

Validation of the Modified Physician‐PREPARED

The validation study for the Modified Physician‐PREPARED occurred in Illinois. The Peoria Institutional Review Board approved and monitored the human research. The patient sample for validation was a prospective cohort from a cluster randomized clinical trial. Willing patients or their proxies provided written consent for study participation. Patient enrollment occurred between December 2004 and August 2006. The subjects for scale analysis were the outpatient primary care physicians or practitioners designated by patients in the cohort. Outpatient physicians and practitioners gave implied consent when they completed and returned questionnaires. Follow‐up was 10 or more days after the patient's discharge from an acute care, 730‐bed, teaching hospital.

Patient Inclusion Criteria

Trained research coordinators identified all consecutive adult inpatients who were discharged to home by internal medicine hospitalist physicians. Patient inclusion in the cluster‐randomized trial required a probability of repeat admission (Pra) score greater than or equal to 0.40.16, 17 Consequently, the patients in the scale analysis cohort had the same high probability for repeat admission. The Pra score came from patient or proxy responses to questions about age, prior hospitalizations, prior doctor visits, self‐rated health status, and other health‐related questions.16, 17 In previous validation studies with elderly outpatients, a Pra score above 0.5 predicted that patients would have 1 hospital admission per person‐year of survival.16 In other validation studies with inpatients aged 18 to 101 years, the Pra items predicted nonroutine discharge planning needs.18

Exclusion Criteria

The exclusion criteria were designed to enroll a cohort with homogeneous risk for readmission. We excluded patients if their discharge destination was a nursing home, another acute care hospital, or an inpatient rehabilitation unit. Hospice patients were excluded if life expectancy was less than 6 months as estimated by the hospitalist. We also used exclusion criteria to avoid illogical enrollments. If the designated outpatient primary care physician or practitioner also managed the patient during the index hospitalization, then there was no perceived barrier to communication and the patient was excluded. Cognitive impairment was a conditional exclusion criterion. We defined cognitive impairment as a score less than 9 on the 10‐point clock test.19 A patient with cognitive impairment could participate with consent from a legally authorized representative. Before we enrolled a cognitively impaired patient, we required a proxy who spent a minimum of 3 hours daily with the patient and who agreed to answer interview questions.

Baseline Assessment

During the index hospitalization, trained data abstractors recorded baseline patient data to calculate the Pra: age, gender, diabetes mellitus, and ischemic heart disease. Patients or proxies provided the number of hospital admissions and doctor visits during the year before the index hospitalization. We recorded the availability of an informal caregiver in response to the question, Is there a friend, relative or neighbor who would take care of you for a few days, if necessary? Patients rated their health status on the following scale: poor, fair, good, very good, and excellent. In addition, we recorded heart failure and chronic obstructive pulmonary disease because of their possible association with readmission.20, 21 Information about outpatient physicians or practitioners came from the hospital's administrative database and was limited to specialty training.

Discharge Process

At the end of the index hospitalization, hospitalists and ward nurses used standardized forms for discharge diagnoses, prescriptions, instructions, and appointments. Discharge planning nurses or social workers consulted with hospitalists and ward nurses and then coordinated service providers including home health nurses, physical therapists, home health aides, homemaker service providers, durable medical equipment vendors, home oxygen vendors, home infusion pharmacists, social workers, rehabilitation service providers, legal aid providers, and others. Patients designated an outpatient primary care physician or nurse practitioner or physician assistant to receive discharge reports and results of diagnostic tests. Ten days after discharge, research personnel mailed the Physician‐PREPARED questionnaire to the designated outpatient primary care professional.

Item Reduction and Scoring

To develop a scale, we selected items from the Physician‐PREPARED survey instrument (see Appendix). Our goal was a parsimonious, comprehensive, and valid scale for use in clinical and research environments. We applied item reduction techniques according to the following steps that were defined a priori. First, we deleted items with nominal response categories that lacked graded or ordinal characteristics. This exclusion criterion caused us to delete the following items from the questionnaire in the appendix: (1a) Who made you aware of the admission, (2a) Who made you aware of the patient's discharge, and (5a) How did you receive this information? We deleted open‐ended questions, such as: (13) Have you any suggestions how the patient's discharge could have been improved? Next, we excluded items with a large proportion of missing responses because respondents checked Not applicable. Only item 12 from the Physician‐PREPARED fulfilled the latter criterion (see Appendix). Question 12 asked, Has the patient's caretaker voiced any concerns that they have not been coping since the patient was discharged? Among 403 respondents, 52% answered question 12 as Not applicable.

Measures of Construct Validity

We used 3 measures of construct validity in our assessment of the Modified Physician‐PREPARED scale. The first construct item asked the outpatient practitioner, Were you involved at all in planning the patient's discharge? The first construct was relevant because involvement by outpatient physicians improves the quality of hospital discharges.22 The second construct item asked, Are you aware of any community support services that are involved in providing assistance to the patient since discharge? For the third construct, we asked (Appendix item 11), Has the patient voiced any concerns that they have not been coping since discharge? We chose community support services and patient coping because these are clinically relevant and correlated with patients' perceptions of discharge preparedness.23 When we assessed construct validity, our hypotheses were significantly higher Modified Physician‐PREPARED scale values for respondents who answered yes to the construct questions about involvement and awareness and answered no to the question about patient‐voiced concerns.

Analysis

Analyses were performed with SPSS PC (version 14.0.2; SPSS Inc, Chicago, Illinois). We reported descriptive statistics as means, standard deviations (SDs), and range for interval variables; median and range for ordinal variables; and percentages for nominal variables. While developing the scale, the unit of analysis was the physician response to a unique patient. Specific descriptive analyses used the unique respondent as the unit of analysis. To determine the internal consistency of the scale, we calculated Cronbach's alpha with SPSS RELIABILITY. We assessed the distribution of the Modified Physician‐PREPARED scale with visual and statistical tests for skewness. While using the SPSS FACTOR program, we performed principal components extractions and then rotated components using the oblique promax technique. Component scores were saved using the regression score procedure. Component loadings above 0.30 were considered for interpretation.24 Statistical inference tests were the Mann‐Whitney U for median differences for 2 groups, the Kruskal‐Wallis for more than 2 groups, and Spearman correlation for associations. The accepted level of significance was P < 0.05.

Description of Validation Cohort for the Modified Physician‐PREPARED

We sent questionnaires to the primary care physician, nurse practitioner, or physician assistant designated by 549 patients. The survey response rate was 76% (417/549). If a respondent failed to check any response option for 2 or more scale items, then the questionnaire was excluded from analysis. We excluded 3% (14/549) of questionnaires for failure to respond to items. The responses from the remaining 403 questionnaires were analyzed. We did not exclude questionnaires from respondents who followed homebound patients or other patients who failed to come to the clinic for postdischarge visits. Our analysis included 90 questionnaires (22%) from respondents who had no contact with the patient after discharge.

The patient characteristics appear in Table 1. Most of the patients were less than 65 years old (77%, 310/403). Many patients had chronic diseases including diabetes mellitus, ischemic heart disease, heart failure, or chronic obstructive pulmonary disease. Most patients, 81% (327/403), rated their health as poor or fair and 55% (223/403) had 1 or more hospital admissions during the year before their index admission. The questionnaire respondents were primary care physicians who practiced internal medicine (41%, 167/403), medicine‐pediatrics (27%, 108/403), family practice (24%, 97/403), or other specialties (3%, 10/403). Nurse practitioners or physician assistants completed 5% (21/403) of questionnaires.

We conducted descriptive analyses that treated the respondent as the unit of analysis. There were 172 unique respondents. The number of questionnaires per respondent ranged from 1 to 20 with a median of 1 questionnaire. Respondents varied in the time to return a questionnaire. We measured response time as the difference between the date we received the questionnaire and the date of discharge. The response time ranged from 10 to 90 days with a median of 21 days after discharge.

Modified Physician‐PREPARED: Item Reduction, Internal Consistency, and Score Distributions

The questionnaire items appear in the Appendix. After item reduction, there were 8 items included in the Modified Physician‐PREPARED scale analysis (Table 2). None of the 8 items caused substantive reduction in Cronbach's alpha, so all were retained. The 8‐item scale had acceptable internal consistency (Cronbach's alpha = 0.86). For an individual questionnaire, the sum of the scores for eight items yielded the Modified Physician‐PREPARED scale value. High scale values reflected high perceptions of discharge quality. Each of the 8 items correlated significantly and positively with the scale value (P < 0.001, 2‐tailed).

Table 2 shows the distribution of responses to each item in the Modified Physician‐PREPARED questionnaire. There were substantial ceiling effects for 2 individual items. One of the 7 items with 3 response options had ceiling effects approaching 70% (item 1). One item had 2 response options and 73% responded yes (item 5). The distribution of Modified Physician‐PREPARED scale values for 403 questionnaires had mean 16.6 4.0 SD and skew 0.6 (standard error of skew = 0.1). When scale values of patients 64 years and younger were compared with those of 65 and older, there were no significant differences (P = 0.606). The scale values did not have noteworthy floor or ceiling effects. The distribution of scale values showed 1.2% (5/403) of respondents had the lowest score of 8 and 1.7% (7/403) had the highest score of 24.

Modified Physician‐PREPARED: Principal Component Analysis

The purpose of the principal component analysis was to evaluate the relationships between the items and domains. In the component analysis, we evaluated the correlation matrix of the 8 items in the Modified Physician‐PREPARED scale. The Kaiser‐Meyer‐Olkin statistic of 0.89 indicated sufficient sampling adequacy to extract components from the matrix. Principal components extracted 66% of the variance associated with the 8‐item scale. After inspection of scree plots, we determined that 2 components were extracted before the eigenvalue fell substantially below 1. The pattern matrix for the promax rotation was inspected and the factor loading for each item appears in Table 3. The item content identified 1 component as timeliness of communication. The other component was adequacy of discharge plan/transmission. Within the adequacy component, the item content addressed patient health status, medication information, and reasons for medication changes. All items loaded primarily on 1 of the components; except item 3, which loaded on both components.

Modified Physician‐PREPARED: Construct Validity

We compared Modified Physician‐PREPARED scale values between dichotomous groups defined by construct variables. When considering the discharge planning and communication for a specific patient, outpatient primary care practitioners reported higher scale values when they were involved in the discharge planning (median [25%, 75%] = 19 [19, 20.5]) than when they were not involved (17 [12.4, 19], P < 0.001). In addition, outpatient practitioners responded with higher scale values when they were aware of community support services (18 [16, 20]) than when they were unaware (17 [12, 19], P = 0.002). There was a nonsignificant trend to higher scale values if patients voiced no concern about coping after discharge (18 [15, 19]) versus concern (17 [12, 19], P = 0.059). For all 3 constructs, the analysis revealed higher Modified Physician‐PREPARED scale values that were in the same direction as hypothesized. We approximated the construct analysis with subscales defined by the principal components (data not shown). The subscale analysis confirmed the direction and significance of the analysis with the full, 8‐item, Modified Physician‐PREPARED scale.

Modified Physician‐PREPARED: Correlations with Baseline Characteristics

We evaluated the correlations between a patient's Modified Physician‐PREPARED scale value and baseline characteristics in Table 1. Patient characteristics were not associated with scale values. We also assessed the median differences between the scale values by practitioner specialty and found no significant differences.

Item Development for the Physician‐PREPARED

Australian investigators designed the Physician‐PREPARED survey instrument to measure the quality of discharge planning activities and communication. The investigators developed the survey with the following process that was not published previously. First, a literature review identified survey content germane to outpatient practitioners.10 Investigators conducted interviews, focus groups, and pilot surveys to prioritize items for the survey instrument. The volunteer subjects for item development were general medical practitioners in Adelaide and Sydney, the capital cities of two states in Australia. The draft instrument was circulated to a small group of general medical practitioners for comment on layout, wording, and question intent. After feedback, minor modifications were made to item content and response categories. The result of development in Australia was a survey instrument with 15 items (see Appendix). The items reflected the following key areas of discharge quality: timeliness of communication, patient health status at discharge, adequacy of discharge support services, discharge medication information, and reasons for medication changes. These areas were congruent with the results of other investigators who assessed the quality of discharge planning and communication.14, 15

Validation of the Modified Physician‐PREPARED

The validation study for the Modified Physician‐PREPARED occurred in Illinois. The Peoria Institutional Review Board approved and monitored the human research. The patient sample for validation was a prospective cohort from a cluster randomized clinical trial. Willing patients or their proxies provided written consent for study participation. Patient enrollment occurred between December 2004 and August 2006. The subjects for scale analysis were the outpatient primary care physicians or practitioners designated by patients in the cohort. Outpatient physicians and practitioners gave implied consent when they completed and returned questionnaires. Follow‐up was 10 or more days after the patient's discharge from an acute care, 730‐bed, teaching hospital.

Patient Inclusion Criteria

Trained research coordinators identified all consecutive adult inpatients who were discharged to home by internal medicine hospitalist physicians. Patient inclusion in the cluster‐randomized trial required a probability of repeat admission (Pra) score greater than or equal to 0.40.16, 17 Consequently, the patients in the scale analysis cohort had the same high probability for repeat admission. The Pra score came from patient or proxy responses to questions about age, prior hospitalizations, prior doctor visits, self‐rated health status, and other health‐related questions.16, 17 In previous validation studies with elderly outpatients, a Pra score above 0.5 predicted that patients would have 1 hospital admission per person‐year of survival.16 In other validation studies with inpatients aged 18 to 101 years, the Pra items predicted nonroutine discharge planning needs.18

Exclusion Criteria

The exclusion criteria were designed to enroll a cohort with homogeneous risk for readmission. We excluded patients if their discharge destination was a nursing home, another acute care hospital, or an inpatient rehabilitation unit. Hospice patients were excluded if life expectancy was less than 6 months as estimated by the hospitalist. We also used exclusion criteria to avoid illogical enrollments. If the designated outpatient primary care physician or practitioner also managed the patient during the index hospitalization, then there was no perceived barrier to communication and the patient was excluded. Cognitive impairment was a conditional exclusion criterion. We defined cognitive impairment as a score less than 9 on the 10‐point clock test.19 A patient with cognitive impairment could participate with consent from a legally authorized representative. Before we enrolled a cognitively impaired patient, we required a proxy who spent a minimum of 3 hours daily with the patient and who agreed to answer interview questions.

Baseline Assessment

During the index hospitalization, trained data abstractors recorded baseline patient data to calculate the Pra: age, gender, diabetes mellitus, and ischemic heart disease. Patients or proxies provided the number of hospital admissions and doctor visits during the year before the index hospitalization. We recorded the availability of an informal caregiver in response to the question, Is there a friend, relative or neighbor who would take care of you for a few days, if necessary? Patients rated their health status on the following scale: poor, fair, good, very good, and excellent. In addition, we recorded heart failure and chronic obstructive pulmonary disease because of their possible association with readmission.20, 21 Information about outpatient physicians or practitioners came from the hospital's administrative database and was limited to specialty training.

Discharge Process

At the end of the index hospitalization, hospitalists and ward nurses used standardized forms for discharge diagnoses, prescriptions, instructions, and appointments. Discharge planning nurses or social workers consulted with hospitalists and ward nurses and then coordinated service providers including home health nurses, physical therapists, home health aides, homemaker service providers, durable medical equipment vendors, home oxygen vendors, home infusion pharmacists, social workers, rehabilitation service providers, legal aid providers, and others. Patients designated an outpatient primary care physician or nurse practitioner or physician assistant to receive discharge reports and results of diagnostic tests. Ten days after discharge, research personnel mailed the Physician‐PREPARED questionnaire to the designated outpatient primary care professional.

Item Reduction and Scoring

To develop a scale, we selected items from the Physician‐PREPARED survey instrument (see Appendix). Our goal was a parsimonious, comprehensive, and valid scale for use in clinical and research environments. We applied item reduction techniques according to the following steps that were defined a priori. First, we deleted items with nominal response categories that lacked graded or ordinal characteristics. This exclusion criterion caused us to delete the following items from the questionnaire in the appendix: (1a) Who made you aware of the admission, (2a) Who made you aware of the patient's discharge, and (5a) How did you receive this information? We deleted open‐ended questions, such as: (13) Have you any suggestions how the patient's discharge could have been improved? Next, we excluded items with a large proportion of missing responses because respondents checked Not applicable. Only item 12 from the Physician‐PREPARED fulfilled the latter criterion (see Appendix). Question 12 asked, Has the patient's caretaker voiced any concerns that they have not been coping since the patient was discharged? Among 403 respondents, 52% answered question 12 as Not applicable.

Measures of Construct Validity

We used 3 measures of construct validity in our assessment of the Modified Physician‐PREPARED scale. The first construct item asked the outpatient practitioner, Were you involved at all in planning the patient's discharge? The first construct was relevant because involvement by outpatient physicians improves the quality of hospital discharges.22 The second construct item asked, Are you aware of any community support services that are involved in providing assistance to the patient since discharge? For the third construct, we asked (Appendix item 11), Has the patient voiced any concerns that they have not been coping since discharge? We chose community support services and patient coping because these are clinically relevant and correlated with patients' perceptions of discharge preparedness.23 When we assessed construct validity, our hypotheses were significantly higher Modified Physician‐PREPARED scale values for respondents who answered yes to the construct questions about involvement and awareness and answered no to the question about patient‐voiced concerns.

Analysis

Analyses were performed with SPSS PC (version 14.0.2; SPSS Inc, Chicago, Illinois). We reported descriptive statistics as means, standard deviations (SDs), and range for interval variables; median and range for ordinal variables; and percentages for nominal variables. While developing the scale, the unit of analysis was the physician response to a unique patient. Specific descriptive analyses used the unique respondent as the unit of analysis. To determine the internal consistency of the scale, we calculated Cronbach's alpha with SPSS RELIABILITY. We assessed the distribution of the Modified Physician‐PREPARED scale with visual and statistical tests for skewness. While using the SPSS FACTOR program, we performed principal components extractions and then rotated components using the oblique promax technique. Component scores were saved using the regression score procedure. Component loadings above 0.30 were considered for interpretation.24 Statistical inference tests were the Mann‐Whitney U for median differences for 2 groups, the Kruskal‐Wallis for more than 2 groups, and Spearman correlation for associations. The accepted level of significance was P < 0.05.

Description of Validation Cohort for the Modified Physician‐PREPARED

We sent questionnaires to the primary care physician, nurse practitioner, or physician assistant designated by 549 patients. The survey response rate was 76% (417/549). If a respondent failed to check any response option for 2 or more scale items, then the questionnaire was excluded from analysis. We excluded 3% (14/549) of questionnaires for failure to respond to items. The responses from the remaining 403 questionnaires were analyzed. We did not exclude questionnaires from respondents who followed homebound patients or other patients who failed to come to the clinic for postdischarge visits. Our analysis included 90 questionnaires (22%) from respondents who had no contact with the patient after discharge.

The patient characteristics appear in Table 1. Most of the patients were less than 65 years old (77%, 310/403). Many patients had chronic diseases including diabetes mellitus, ischemic heart disease, heart failure, or chronic obstructive pulmonary disease. Most patients, 81% (327/403), rated their health as poor or fair and 55% (223/403) had 1 or more hospital admissions during the year before their index admission. The questionnaire respondents were primary care physicians who practiced internal medicine (41%, 167/403), medicine‐pediatrics (27%, 108/403), family practice (24%, 97/403), or other specialties (3%, 10/403). Nurse practitioners or physician assistants completed 5% (21/403) of questionnaires.

We conducted descriptive analyses that treated the respondent as the unit of analysis. There were 172 unique respondents. The number of questionnaires per respondent ranged from 1 to 20 with a median of 1 questionnaire. Respondents varied in the time to return a questionnaire. We measured response time as the difference between the date we received the questionnaire and the date of discharge. The response time ranged from 10 to 90 days with a median of 21 days after discharge.

Modified Physician‐PREPARED: Item Reduction, Internal Consistency, and Score Distributions

The questionnaire items appear in the Appendix. After item reduction, there were 8 items included in the Modified Physician‐PREPARED scale analysis (Table 2). None of the 8 items caused substantive reduction in Cronbach's alpha, so all were retained. The 8‐item scale had acceptable internal consistency (Cronbach's alpha = 0.86). For an individual questionnaire, the sum of the scores for eight items yielded the Modified Physician‐PREPARED scale value. High scale values reflected high perceptions of discharge quality. Each of the 8 items correlated significantly and positively with the scale value (P < 0.001, 2‐tailed).

Table 2 shows the distribution of responses to each item in the Modified Physician‐PREPARED questionnaire. There were substantial ceiling effects for 2 individual items. One of the 7 items with 3 response options had ceiling effects approaching 70% (item 1). One item had 2 response options and 73% responded yes (item 5). The distribution of Modified Physician‐PREPARED scale values for 403 questionnaires had mean 16.6 4.0 SD and skew 0.6 (standard error of skew = 0.1). When scale values of patients 64 years and younger were compared with those of 65 and older, there were no significant differences (P = 0.606). The scale values did not have noteworthy floor or ceiling effects. The distribution of scale values showed 1.2% (5/403) of respondents had the lowest score of 8 and 1.7% (7/403) had the highest score of 24.

Modified Physician‐PREPARED: Principal Component Analysis

The purpose of the principal component analysis was to evaluate the relationships between the items and domains. In the component analysis, we evaluated the correlation matrix of the 8 items in the Modified Physician‐PREPARED scale. The Kaiser‐Meyer‐Olkin statistic of 0.89 indicated sufficient sampling adequacy to extract components from the matrix. Principal components extracted 66% of the variance associated with the 8‐item scale. After inspection of scree plots, we determined that 2 components were extracted before the eigenvalue fell substantially below 1. The pattern matrix for the promax rotation was inspected and the factor loading for each item appears in Table 3. The item content identified 1 component as timeliness of communication. The other component was adequacy of discharge plan/transmission. Within the adequacy component, the item content addressed patient health status, medication information, and reasons for medication changes. All items loaded primarily on 1 of the components; except item 3, which loaded on both components.

Modified Physician‐PREPARED: Construct Validity

We compared Modified Physician‐PREPARED scale values between dichotomous groups defined by construct variables. When considering the discharge planning and communication for a specific patient, outpatient primary care practitioners reported higher scale values when they were involved in the discharge planning (median [25%, 75%] = 19 [19, 20.5]) than when they were not involved (17 [12.4, 19], P < 0.001). In addition, outpatient practitioners responded with higher scale values when they were aware of community support services (18 [16, 20]) than when they were unaware (17 [12, 19], P = 0.002). There was a nonsignificant trend to higher scale values if patients voiced no concern about coping after discharge (18 [15, 19]) versus concern (17 [12, 19], P = 0.059). For all 3 constructs, the analysis revealed higher Modified Physician‐PREPARED scale values that were in the same direction as hypothesized. We approximated the construct analysis with subscales defined by the principal components (data not shown). The subscale analysis confirmed the direction and significance of the analysis with the full, 8‐item, Modified Physician‐PREPARED scale.

Modified Physician‐PREPARED: Correlations with Baseline Characteristics

We evaluated the correlations between a patient's Modified Physician‐PREPARED scale value and baseline characteristics in Table 1. Patient characteristics were not associated with scale values. We also assessed the median differences between the scale values by practitioner specialty and found no significant differences.

Item Development for the Physician‐PREPARED

Australian investigators designed the Physician‐PREPARED survey instrument to measure the quality of discharge planning activities and communication. The investigators developed the survey with the following process that was not published previously. First, a literature review identified survey content germane to outpatient practitioners.10 Investigators conducted interviews, focus groups, and pilot surveys to prioritize items for the survey instrument. The volunteer subjects for item development were general medical practitioners in Adelaide and Sydney, the capital cities of two states in Australia. The draft instrument was circulated to a small group of general medical practitioners for comment on layout, wording, and question intent. After feedback, minor modifications were made to item content and response categories. The result of development in Australia was a survey instrument with 15 items (see Appendix). The items reflected the following key areas of discharge quality: timeliness of communication, patient health status at discharge, adequacy of discharge support services, discharge medication information, and reasons for medication changes. These areas were congruent with the results of other investigators who assessed the quality of discharge planning and communication.14, 15

Validation of the Modified Physician‐PREPARED

The validation study for the Modified Physician‐PREPARED occurred in Illinois. The Peoria Institutional Review Board approved and monitored the human research. The patient sample for validation was a prospective cohort from a cluster randomized clinical trial. Willing patients or their proxies provided written consent for study participation. Patient enrollment occurred between December 2004 and August 2006. The subjects for scale analysis were the outpatient primary care physicians or practitioners designated by patients in the cohort. Outpatient physicians and practitioners gave implied consent when they completed and returned questionnaires. Follow‐up was 10 or more days after the patient's discharge from an acute care, 730‐bed, teaching hospital.

Patient Inclusion Criteria

Trained research coordinators identified all consecutive adult inpatients who were discharged to home by internal medicine hospitalist physicians. Patient inclusion in the cluster‐randomized trial required a probability of repeat admission (Pra) score greater than or equal to 0.40.16, 17 Consequently, the patients in the scale analysis cohort had the same high probability for repeat admission. The Pra score came from patient or proxy responses to questions about age, prior hospitalizations, prior doctor visits, self‐rated health status, and other health‐related questions.16, 17 In previous validation studies with elderly outpatients, a Pra score above 0.5 predicted that patients would have 1 hospital admission per person‐year of survival.16 In other validation studies with inpatients aged 18 to 101 years, the Pra items predicted nonroutine discharge planning needs.18

Exclusion Criteria

The exclusion criteria were designed to enroll a cohort with homogeneous risk for readmission. We excluded patients if their discharge destination was a nursing home, another acute care hospital, or an inpatient rehabilitation unit. Hospice patients were excluded if life expectancy was less than 6 months as estimated by the hospitalist. We also used exclusion criteria to avoid illogical enrollments. If the designated outpatient primary care physician or practitioner also managed the patient during the index hospitalization, then there was no perceived barrier to communication and the patient was excluded. Cognitive impairment was a conditional exclusion criterion. We defined cognitive impairment as a score less than 9 on the 10‐point clock test.19 A patient with cognitive impairment could participate with consent from a legally authorized representative. Before we enrolled a cognitively impaired patient, we required a proxy who spent a minimum of 3 hours daily with the patient and who agreed to answer interview questions.

Baseline Assessment

During the index hospitalization, trained data abstractors recorded baseline patient data to calculate the Pra: age, gender, diabetes mellitus, and ischemic heart disease. Patients or proxies provided the number of hospital admissions and doctor visits during the year before the index hospitalization. We recorded the availability of an informal caregiver in response to the question, Is there a friend, relative or neighbor who would take care of you for a few days, if necessary? Patients rated their health status on the following scale: poor, fair, good, very good, and excellent. In addition, we recorded heart failure and chronic obstructive pulmonary disease because of their possible association with readmission.20, 21 Information about outpatient physicians or practitioners came from the hospital's administrative database and was limited to specialty training.

Discharge Process

At the end of the index hospitalization, hospitalists and ward nurses used standardized forms for discharge diagnoses, prescriptions, instructions, and appointments. Discharge planning nurses or social workers consulted with hospitalists and ward nurses and then coordinated service providers including home health nurses, physical therapists, home health aides, homemaker service providers, durable medical equipment vendors, home oxygen vendors, home infusion pharmacists, social workers, rehabilitation service providers, legal aid providers, and others. Patients designated an outpatient primary care physician or nurse practitioner or physician assistant to receive discharge reports and results of diagnostic tests. Ten days after discharge, research personnel mailed the Physician‐PREPARED questionnaire to the designated outpatient primary care professional.

Item Reduction and Scoring

To develop a scale, we selected items from the Physician‐PREPARED survey instrument (see Appendix). Our goal was a parsimonious, comprehensive, and valid scale for use in clinical and research environments. We applied item reduction techniques according to the following steps that were defined a priori. First, we deleted items with nominal response categories that lacked graded or ordinal characteristics. This exclusion criterion caused us to delete the following items from the questionnaire in the appendix: (1a) Who made you aware of the admission, (2a) Who made you aware of the patient's discharge, and (5a) How did you receive this information? We deleted open‐ended questions, such as: (13) Have you any suggestions how the patient's discharge could have been improved? Next, we excluded items with a large proportion of missing responses because respondents checked Not applicable. Only item 12 from the Physician‐PREPARED fulfilled the latter criterion (see Appendix). Question 12 asked, Has the patient's caretaker voiced any concerns that they have not been coping since the patient was discharged? Among 403 respondents, 52% answered question 12 as Not applicable.

Measures of Construct Validity

We used 3 measures of construct validity in our assessment of the Modified Physician‐PREPARED scale. The first construct item asked the outpatient practitioner, Were you involved at all in planning the patient's discharge? The first construct was relevant because involvement by outpatient physicians improves the quality of hospital discharges.22 The second construct item asked, Are you aware of any community support services that are involved in providing assistance to the patient since discharge? For the third construct, we asked (Appendix item 11), Has the patient voiced any concerns that they have not been coping since discharge? We chose community support services and patient coping because these are clinically relevant and correlated with patients' perceptions of discharge preparedness.23 When we assessed construct validity, our hypotheses were significantly higher Modified Physician‐PREPARED scale values for respondents who answered yes to the construct questions about involvement and awareness and answered no to the question about patient‐voiced concerns.

Analysis

Analyses were performed with SPSS PC (version 14.0.2; SPSS Inc, Chicago, Illinois). We reported descriptive statistics as means, standard deviations (SDs), and range for interval variables; median and range for ordinal variables; and percentages for nominal variables. While developing the scale, the unit of analysis was the physician response to a unique patient. Specific descriptive analyses used the unique respondent as the unit of analysis. To determine the internal consistency of the scale, we calculated Cronbach's alpha with SPSS RELIABILITY. We assessed the distribution of the Modified Physician‐PREPARED scale with visual and statistical tests for skewness. While using the SPSS FACTOR program, we performed principal components extractions and then rotated components using the oblique promax technique. Component scores were saved using the regression score procedure. Component loadings above 0.30 were considered for interpretation.24 Statistical inference tests were the Mann‐Whitney U for median differences for 2 groups, the Kruskal‐Wallis for more than 2 groups, and Spearman correlation for associations. The accepted level of significance was P < 0.05.

Description of Validation Cohort for the Modified Physician‐PREPARED

We sent questionnaires to the primary care physician, nurse practitioner, or physician assistant designated by 549 patients. The survey response rate was 76% (417/549). If a respondent failed to check any response option for 2 or more scale items, then the questionnaire was excluded from analysis. We excluded 3% (14/549) of questionnaires for failure to respond to items. The responses from the remaining 403 questionnaires were analyzed. We did not exclude questionnaires from respondents who followed homebound patients or other patients who failed to come to the clinic for postdischarge visits. Our analysis included 90 questionnaires (22%) from respondents who had no contact with the patient after discharge.

The patient characteristics appear in Table 1. Most of the patients were less than 65 years old (77%, 310/403). Many patients had chronic diseases including diabetes mellitus, ischemic heart disease, heart failure, or chronic obstructive pulmonary disease. Most patients, 81% (327/403), rated their health as poor or fair and 55% (223/403) had 1 or more hospital admissions during the year before their index admission. The questionnaire respondents were primary care physicians who practiced internal medicine (41%, 167/403), medicine‐pediatrics (27%, 108/403), family practice (24%, 97/403), or other specialties (3%, 10/403). Nurse practitioners or physician assistants completed 5% (21/403) of questionnaires.

We conducted descriptive analyses that treated the respondent as the unit of analysis. There were 172 unique respondents. The number of questionnaires per respondent ranged from 1 to 20 with a median of 1 questionnaire. Respondents varied in the time to return a questionnaire. We measured response time as the difference between the date we received the questionnaire and the date of discharge. The response time ranged from 10 to 90 days with a median of 21 days after discharge.

Modified Physician‐PREPARED: Item Reduction, Internal Consistency, and Score Distributions

The questionnaire items appear in the Appendix. After item reduction, there were 8 items included in the Modified Physician‐PREPARED scale analysis (Table 2). None of the 8 items caused substantive reduction in Cronbach's alpha, so all were retained. The 8‐item scale had acceptable internal consistency (Cronbach's alpha = 0.86). For an individual questionnaire, the sum of the scores for eight items yielded the Modified Physician‐PREPARED scale value. High scale values reflected high perceptions of discharge quality. Each of the 8 items correlated significantly and positively with the scale value (P < 0.001, 2‐tailed).

Table 2 shows the distribution of responses to each item in the Modified Physician‐PREPARED questionnaire. There were substantial ceiling effects for 2 individual items. One of the 7 items with 3 response options had ceiling effects approaching 70% (item 1). One item had 2 response options and 73% responded yes (item 5). The distribution of Modified Physician‐PREPARED scale values for 403 questionnaires had mean 16.6 4.0 SD and skew 0.6 (standard error of skew = 0.1). When scale values of patients 64 years and younger were compared with those of 65 and older, there were no significant differences (P = 0.606). The scale values did not have noteworthy floor or ceiling effects. The distribution of scale values showed 1.2% (5/403) of respondents had the lowest score of 8 and 1.7% (7/403) had the highest score of 24.

Modified Physician‐PREPARED: Principal Component Analysis

The purpose of the principal component analysis was to evaluate the relationships between the items and domains. In the component analysis, we evaluated the correlation matrix of the 8 items in the Modified Physician‐PREPARED scale. The Kaiser‐Meyer‐Olkin statistic of 0.89 indicated sufficient sampling adequacy to extract components from the matrix. Principal components extracted 66% of the variance associated with the 8‐item scale. After inspection of scree plots, we determined that 2 components were extracted before the eigenvalue fell substantially below 1. The pattern matrix for the promax rotation was inspected and the factor loading for each item appears in Table 3. The item content identified 1 component as timeliness of communication. The other component was adequacy of discharge plan/transmission. Within the adequacy component, the item content addressed patient health status, medication information, and reasons for medication changes. All items loaded primarily on 1 of the components; except item 3, which loaded on both components.

Modified Physician‐PREPARED: Construct Validity

We compared Modified Physician‐PREPARED scale values between dichotomous groups defined by construct variables. When considering the discharge planning and communication for a specific patient, outpatient primary care practitioners reported higher scale values when they were involved in the discharge planning (median [25%, 75%] = 19 [19, 20.5]) than when they were not involved (17 [12.4, 19], P < 0.001). In addition, outpatient practitioners responded with higher scale values when they were aware of community support services (18 [16, 20]) than when they were unaware (17 [12, 19], P = 0.002). There was a nonsignificant trend to higher scale values if patients voiced no concern about coping after discharge (18 [15, 19]) versus concern (17 [12, 19], P = 0.059). For all 3 constructs, the analysis revealed higher Modified Physician‐PREPARED scale values that were in the same direction as hypothesized. We approximated the construct analysis with subscales defined by the principal components (data not shown). The subscale analysis confirmed the direction and significance of the analysis with the full, 8‐item, Modified Physician‐PREPARED scale.

Modified Physician‐PREPARED: Correlations with Baseline Characteristics

We evaluated the correlations between a patient's Modified Physician‐PREPARED scale value and baseline characteristics in Table 1. Patient characteristics were not associated with scale values. We also assessed the median differences between the scale values by practitioner specialty and found no significant differences.

Sites

The Multicenter Hospitalist Study (MCHS) was a multicenter trial of general medical services that enrolled patients at 6 geographically diverse centers: The University of Chicago (which also served as the coordinating center), University of Iowa Hospitals and Clinics, University of California San Francisco, University of Wisconsin, University of New Mexico, and Brigham and Women's Hospital.19

Each site was selected to participate in the MCHS because patients on their general medicine service were admitted to hospitalist and nonhospitalist physicians in a random fashion (eg, based on predetermined call schedule based on day of the week). As teaching hospitals, house officers provided direct care to patients hospitalized at each center; nonteaching services were not present at the sites during the period of this study.

During the period of this study, each site complied with PSDA requirements for noting that patients had been informed about their right to create an advance directive, but no sites had a guideline or other program in place specifically intended to facilitate physician‐patient communication about care wishes. Two sites had active Hospice or Palliative Care services, and another 2 had Geriatrics Consultation services, but none had standard protocols mandating involvement of these consultants at the time of admission, the time when our key outcomes were documented.

Patients

Patients were eligible for inclusion in the MCHS if they were older than 18 years of age and were admitted at random to a hospitalist or nonhospitalist physician; we excluded patients from MCHS if they were admitted specifically under the care of their primary care physician or subspecialist (eg, admitted for chemotherapy) or were a prison inmate. Patients meeting these eligibility criteria were then approached for purposes of informed consent.

Data Collection

Data for this study were obtained from administrative data, patient interview, and chart abstraction as in previous work.14 Administrative data were drawn from cost‐accounting databases at each participating hospital; administrative data were used to provide cost and length of stay data, as well as information about patient insurance type, age, and sex.

We interviewed patients immediately after informed consent was obtained, with both taking place generally within 24 hours of admission. Interviews collected data about patient preferences for care and functional status,20 and other data not reliably available from administrative sources (such as housing situation).

Patient care plan before admission was taken from notes and orders written in the first 24 hours of hospitalization, as mentioned above. Using criteria we employed in previous work,21 a care discussion (CD) was defined as documentation of a discussion between patients (or family) and at least 1 physician (primary physician, hospitalist, consulting physician, or house officer) during the first 24 hours of hospitalization. CDs needed to specify that the person who wrote the note had actually spoken with the patient or their family for the purposes of determining preferences for care, and that this discussion resulted in a specific care plan. Thus, notations such as do not resuscitate/do not intubate, or spoke with family, questions answered, did not qualify as CDs, but a note stating the patient continues to want full efforts was counted as a CD.

Principal investigators at each site were responsible for training and overseeing interviewing and chart abstraction activities at each site, with central oversight of data quality provided by the central coordinating center. Upon receipt at the data coordinating center, all data were examined for missing, nonsensical, or outlier data with errors referred back to the participating sites for correction.

Statistical Analysis

For bivariable comparisons of patients with and without CDs, we used chi‐squared or Mann‐Whitney U‐tests, as appropriate.

Variables with P < 0.20 in bivariable comparisons were selected for initial inclusion in models. Then, using automated forward and stepwise selection techniques as well as manually entered variables, we fit multivariable generalized estimating equations permitting clustering of effects at the physician level to determine the independent association between the multiple factors tested and presence of a CD. In order to guard against the threat of multiple testing, we retained variables at a significance level of P < 0.01; variables were also retained because of observed confounding with other independent variables, or to maintain face validity of the model. All analyses were performed using SAS 9.0 for Windows (SAS Institute Inc., Cary, NC).

Patient Sociodemographics (Table 1)

A total of 17,097 of 33,638 patients (50.8%) were interviewed and gave consent for chart abstraction. Of these patients, 1776 (10.3%) had a CD documented in the first 24 hours of hospitalization. Patients with documented CDs were older, more often white, had completed more years of education, were more likely to have lived in a nursing home prior to admission, and more likely to have been hospitalized in the last 12 months. The proportion of patients with CDs was highly variable across site of enrollment, from 2.8%‐24.9%.

Patient Self‐Reported Health Status and Comorbid Illness (Table 2)

Patients with CDs more often reported a lot of difficulties with bathing, eating, or dressing; household chores; and moderate activities. Patients with CDs were more likely to report accomplishing less than they would like due to their health. They were more likely to have cancer, depression, a history of stroke, and heart disease, but less likely to have diabetes or human immunodeficiency virus.

Patient Preferences, Care Plan Documentation, and Care Coordination at Admission (Table 3)

Patients who had documented CDs were less likely to prefer my doctor give me choices regarding my care, and more often disagreed with the statement I prefer to leave care decisions to my physician. These patients were also more likely to have a durable power of attorney or living will in their chart, or have an alternate decision‐maker noted. The majority of patients without a documented CD (79.9%) had no notation of their care wishes, compared to 29.7% in patients with a documented CD. Patients with a documented CD were more likely to have a regular medical provider and a note in the chart from their primary care physician.

Factors Associated with Documented Care Discussions (Table 4)

Using predictor variables presented in Tables 1‐3, we then constructed multivariable models seeking to understand factors independently associated with documentation of code status in the entire cohort, as well as among patients who had no preexisting care wishes.

In the entire cohort, insurance type was independently associated with likelihood of a care discussion, with patients with Medicare having greater adjusted odds ratio for a CD than patients with all other forms of insurance, even after adjusting for age. Patients who had functional limitations with bathing, toileting, and feeding; had a documented surrogate decision maker; were unable to participate in their care; had cancer; or did not live in their own home were more likely to have a documented CD. Subjects with diabetes were less likely to have a CD, although this was of borderline significance. Patients whose team had documented a CD with the patients' primary physician were also more likely to have a discussion noted. However, the magnitude of these predictors was small compared to the independent effects attributable to the site the patient was enrolled or whether the patient had any preexisting documentation. Whereas the adjusted odds ratio associated with clinical or functional measures (such as age, cancer) were generally between 1.5 and 2.5, the range of odds ratios associated with having any documentation of care wishes (compared to no documentation) were all greater than 3, and the odds ratios associated with site of enrollment were 1.7 or higher.

We observed similar findings in analyses limited to patients with no preexisting care documentation. While clinical, sociodemographic, and functional factors remained statistically associated with a CD (albeit with wider confidence intervals due to smaller sample sizes), the effect of the patient's site of enrollment became even more striking (Table 4).

Sites

The Multicenter Hospitalist Study (MCHS) was a multicenter trial of general medical services that enrolled patients at 6 geographically diverse centers: The University of Chicago (which also served as the coordinating center), University of Iowa Hospitals and Clinics, University of California San Francisco, University of Wisconsin, University of New Mexico, and Brigham and Women's Hospital.19

Each site was selected to participate in the MCHS because patients on their general medicine service were admitted to hospitalist and nonhospitalist physicians in a random fashion (eg, based on predetermined call schedule based on day of the week). As teaching hospitals, house officers provided direct care to patients hospitalized at each center; nonteaching services were not present at the sites during the period of this study.

During the period of this study, each site complied with PSDA requirements for noting that patients had been informed about their right to create an advance directive, but no sites had a guideline or other program in place specifically intended to facilitate physician‐patient communication about care wishes. Two sites had active Hospice or Palliative Care services, and another 2 had Geriatrics Consultation services, but none had standard protocols mandating involvement of these consultants at the time of admission, the time when our key outcomes were documented.

Patients

Patients were eligible for inclusion in the MCHS if they were older than 18 years of age and were admitted at random to a hospitalist or nonhospitalist physician; we excluded patients from MCHS if they were admitted specifically under the care of their primary care physician or subspecialist (eg, admitted for chemotherapy) or were a prison inmate. Patients meeting these eligibility criteria were then approached for purposes of informed consent.

Data Collection

Data for this study were obtained from administrative data, patient interview, and chart abstraction as in previous work.14 Administrative data were drawn from cost‐accounting databases at each participating hospital; administrative data were used to provide cost and length of stay data, as well as information about patient insurance type, age, and sex.

We interviewed patients immediately after informed consent was obtained, with both taking place generally within 24 hours of admission. Interviews collected data about patient preferences for care and functional status,20 and other data not reliably available from administrative sources (such as housing situation).

Patient care plan before admission was taken from notes and orders written in the first 24 hours of hospitalization, as mentioned above. Using criteria we employed in previous work,21 a care discussion (CD) was defined as documentation of a discussion between patients (or family) and at least 1 physician (primary physician, hospitalist, consulting physician, or house officer) during the first 24 hours of hospitalization. CDs needed to specify that the person who wrote the note had actually spoken with the patient or their family for the purposes of determining preferences for care, and that this discussion resulted in a specific care plan. Thus, notations such as do not resuscitate/do not intubate, or spoke with family, questions answered, did not qualify as CDs, but a note stating the patient continues to want full efforts was counted as a CD.

Principal investigators at each site were responsible for training and overseeing interviewing and chart abstraction activities at each site, with central oversight of data quality provided by the central coordinating center. Upon receipt at the data coordinating center, all data were examined for missing, nonsensical, or outlier data with errors referred back to the participating sites for correction.

Statistical Analysis

For bivariable comparisons of patients with and without CDs, we used chi‐squared or Mann‐Whitney U‐tests, as appropriate.

Variables with P < 0.20 in bivariable comparisons were selected for initial inclusion in models. Then, using automated forward and stepwise selection techniques as well as manually entered variables, we fit multivariable generalized estimating equations permitting clustering of effects at the physician level to determine the independent association between the multiple factors tested and presence of a CD. In order to guard against the threat of multiple testing, we retained variables at a significance level of P < 0.01; variables were also retained because of observed confounding with other independent variables, or to maintain face validity of the model. All analyses were performed using SAS 9.0 for Windows (SAS Institute Inc., Cary, NC).

Patient Sociodemographics (Table 1)

A total of 17,097 of 33,638 patients (50.8%) were interviewed and gave consent for chart abstraction. Of these patients, 1776 (10.3%) had a CD documented in the first 24 hours of hospitalization. Patients with documented CDs were older, more often white, had completed more years of education, were more likely to have lived in a nursing home prior to admission, and more likely to have been hospitalized in the last 12 months. The proportion of patients with CDs was highly variable across site of enrollment, from 2.8%‐24.9%.

Patient Self‐Reported Health Status and Comorbid Illness (Table 2)

Patients with CDs more often reported a lot of difficulties with bathing, eating, or dressing; household chores; and moderate activities. Patients with CDs were more likely to report accomplishing less than they would like due to their health. They were more likely to have cancer, depression, a history of stroke, and heart disease, but less likely to have diabetes or human immunodeficiency virus.

Patient Preferences, Care Plan Documentation, and Care Coordination at Admission (Table 3)

Patients who had documented CDs were less likely to prefer my doctor give me choices regarding my care, and more often disagreed with the statement I prefer to leave care decisions to my physician. These patients were also more likely to have a durable power of attorney or living will in their chart, or have an alternate decision‐maker noted. The majority of patients without a documented CD (79.9%) had no notation of their care wishes, compared to 29.7% in patients with a documented CD. Patients with a documented CD were more likely to have a regular medical provider and a note in the chart from their primary care physician.

Factors Associated with Documented Care Discussions (Table 4)

Using predictor variables presented in Tables 1‐3, we then constructed multivariable models seeking to understand factors independently associated with documentation of code status in the entire cohort, as well as among patients who had no preexisting care wishes.

In the entire cohort, insurance type was independently associated with likelihood of a care discussion, with patients with Medicare having greater adjusted odds ratio for a CD than patients with all other forms of insurance, even after adjusting for age. Patients who had functional limitations with bathing, toileting, and feeding; had a documented surrogate decision maker; were unable to participate in their care; had cancer; or did not live in their own home were more likely to have a documented CD. Subjects with diabetes were less likely to have a CD, although this was of borderline significance. Patients whose team had documented a CD with the patients' primary physician were also more likely to have a discussion noted. However, the magnitude of these predictors was small compared to the independent effects attributable to the site the patient was enrolled or whether the patient had any preexisting documentation. Whereas the adjusted odds ratio associated with clinical or functional measures (such as age, cancer) were generally between 1.5 and 2.5, the range of odds ratios associated with having any documentation of care wishes (compared to no documentation) were all greater than 3, and the odds ratios associated with site of enrollment were 1.7 or higher.

We observed similar findings in analyses limited to patients with no preexisting care documentation. While clinical, sociodemographic, and functional factors remained statistically associated with a CD (albeit with wider confidence intervals due to smaller sample sizes), the effect of the patient's site of enrollment became even more striking (Table 4).

Sites

The Multicenter Hospitalist Study (MCHS) was a multicenter trial of general medical services that enrolled patients at 6 geographically diverse centers: The University of Chicago (which also served as the coordinating center), University of Iowa Hospitals and Clinics, University of California San Francisco, University of Wisconsin, University of New Mexico, and Brigham and Women's Hospital.19

Each site was selected to participate in the MCHS because patients on their general medicine service were admitted to hospitalist and nonhospitalist physicians in a random fashion (eg, based on predetermined call schedule based on day of the week). As teaching hospitals, house officers provided direct care to patients hospitalized at each center; nonteaching services were not present at the sites during the period of this study.

During the period of this study, each site complied with PSDA requirements for noting that patients had been informed about their right to create an advance directive, but no sites had a guideline or other program in place specifically intended to facilitate physician‐patient communication about care wishes. Two sites had active Hospice or Palliative Care services, and another 2 had Geriatrics Consultation services, but none had standard protocols mandating involvement of these consultants at the time of admission, the time when our key outcomes were documented.

Patients

Patients were eligible for inclusion in the MCHS if they were older than 18 years of age and were admitted at random to a hospitalist or nonhospitalist physician; we excluded patients from MCHS if they were admitted specifically under the care of their primary care physician or subspecialist (eg, admitted for chemotherapy) or were a prison inmate. Patients meeting these eligibility criteria were then approached for purposes of informed consent.

Data Collection

Data for this study were obtained from administrative data, patient interview, and chart abstraction as in previous work.14 Administrative data were drawn from cost‐accounting databases at each participating hospital; administrative data were used to provide cost and length of stay data, as well as information about patient insurance type, age, and sex.

We interviewed patients immediately after informed consent was obtained, with both taking place generally within 24 hours of admission. Interviews collected data about patient preferences for care and functional status,20 and other data not reliably available from administrative sources (such as housing situation).

Patient care plan before admission was taken from notes and orders written in the first 24 hours of hospitalization, as mentioned above. Using criteria we employed in previous work,21 a care discussion (CD) was defined as documentation of a discussion between patients (or family) and at least 1 physician (primary physician, hospitalist, consulting physician, or house officer) during the first 24 hours of hospitalization. CDs needed to specify that the person who wrote the note had actually spoken with the patient or their family for the purposes of determining preferences for care, and that this discussion resulted in a specific care plan. Thus, notations such as do not resuscitate/do not intubate, or spoke with family, questions answered, did not qualify as CDs, but a note stating the patient continues to want full efforts was counted as a CD.

Principal investigators at each site were responsible for training and overseeing interviewing and chart abstraction activities at each site, with central oversight of data quality provided by the central coordinating center. Upon receipt at the data coordinating center, all data were examined for missing, nonsensical, or outlier data with errors referred back to the participating sites for correction.

Statistical Analysis

For bivariable comparisons of patients with and without CDs, we used chi‐squared or Mann‐Whitney U‐tests, as appropriate.

Variables with P < 0.20 in bivariable comparisons were selected for initial inclusion in models. Then, using automated forward and stepwise selection techniques as well as manually entered variables, we fit multivariable generalized estimating equations permitting clustering of effects at the physician level to determine the independent association between the multiple factors tested and presence of a CD. In order to guard against the threat of multiple testing, we retained variables at a significance level of P < 0.01; variables were also retained because of observed confounding with other independent variables, or to maintain face validity of the model. All analyses were performed using SAS 9.0 for Windows (SAS Institute Inc., Cary, NC).

Patient Sociodemographics (Table 1)

A total of 17,097 of 33,638 patients (50.8%) were interviewed and gave consent for chart abstraction. Of these patients, 1776 (10.3%) had a CD documented in the first 24 hours of hospitalization. Patients with documented CDs were older, more often white, had completed more years of education, were more likely to have lived in a nursing home prior to admission, and more likely to have been hospitalized in the last 12 months. The proportion of patients with CDs was highly variable across site of enrollment, from 2.8%‐24.9%.

Patient Self‐Reported Health Status and Comorbid Illness (Table 2)

Patients with CDs more often reported a lot of difficulties with bathing, eating, or dressing; household chores; and moderate activities. Patients with CDs were more likely to report accomplishing less than they would like due to their health. They were more likely to have cancer, depression, a history of stroke, and heart disease, but less likely to have diabetes or human immunodeficiency virus.

Patient Preferences, Care Plan Documentation, and Care Coordination at Admission (Table 3)

Patients who had documented CDs were less likely to prefer my doctor give me choices regarding my care, and more often disagreed with the statement I prefer to leave care decisions to my physician. These patients were also more likely to have a durable power of attorney or living will in their chart, or have an alternate decision‐maker noted. The majority of patients without a documented CD (79.9%) had no notation of their care wishes, compared to 29.7% in patients with a documented CD. Patients with a documented CD were more likely to have a regular medical provider and a note in the chart from their primary care physician.

Factors Associated with Documented Care Discussions (Table 4)

Using predictor variables presented in Tables 1‐3, we then constructed multivariable models seeking to understand factors independently associated with documentation of code status in the entire cohort, as well as among patients who had no preexisting care wishes.

In the entire cohort, insurance type was independently associated with likelihood of a care discussion, with patients with Medicare having greater adjusted odds ratio for a CD than patients with all other forms of insurance, even after adjusting for age. Patients who had functional limitations with bathing, toileting, and feeding; had a documented surrogate decision maker; were unable to participate in their care; had cancer; or did not live in their own home were more likely to have a documented CD. Subjects with diabetes were less likely to have a CD, although this was of borderline significance. Patients whose team had documented a CD with the patients' primary physician were also more likely to have a discussion noted. However, the magnitude of these predictors was small compared to the independent effects attributable to the site the patient was enrolled or whether the patient had any preexisting documentation. Whereas the adjusted odds ratio associated with clinical or functional measures (such as age, cancer) were generally between 1.5 and 2.5, the range of odds ratios associated with having any documentation of care wishes (compared to no documentation) were all greater than 3, and the odds ratios associated with site of enrollment were 1.7 or higher.

We observed similar findings in analyses limited to patients with no preexisting care documentation. While clinical, sociodemographic, and functional factors remained statistically associated with a CD (albeit with wider confidence intervals due to smaller sample sizes), the effect of the patient's site of enrollment became even more striking (Table 4).

General Medical Disorders

Primary sleep disorders, such as obstructive sleep apnea (OSA) and numerous other medical illnesses, can directly impair sleep physiology, leading to a cyclical interaction (Figure 2). Other conditions that disrupt sleep include congestive heart failure (CHF), diabetes mellitus, chronic obstructive pulmonary disease (COPD), gastroesophageal reflux, cardiovascular disease, thyroid disorders, renal disease, and severe liver disease.26 Table 2 lists selected medical and neurological conditions, their associated sleep‐related problems, and suggestions on how to ameliorate these problems.

Figure 2

Affecting approximately 24% of men and 9% of women in the United States, OSA is the most common primary sleep disorder,27, 28 and causes significant mental and physical morbidity. Risk factors for OSA include obesity, hypothyroidism‐induced muscle weakness, and structural abnormalities in the oropharynx region such as acromegaly, micrognathia, or retrognathia. OSA is characterized by episodes of complete or partial pharyngeal obstruction during sleep that cause snoring, apneic episodes, choking, dyspnea, and restlessness.28 These episodes are associated with intermittent nocturnal sympathetic activation leading to nocturnal awakenings and cortical arousals, all of which lead to daytime symptoms of fatigue, sleepiness, and cognitive impairment (Figure 2). In addition, chronic sympathetic activation causes numerous derangements in the vascular endothelium and platelet activation.29, 30 Sleep‐disordered breathing has been independently associated with cardiovascular diseases such as hypertension, CHF, ischemic heart disease, atrial fibrillation, and cerebrovascular disease.31, 32

OSA is also associated with sleep‐related gastroesophageal reflux, which is characterized by pain and nocturnal cough, and can induce nocturnal asthma attacks and laryngospasm.33 Green et al.29 found that OSA patients treated with continuous positive airway pressure (CPAP) had a 48% improvement in nocturnal reflux symptoms. Although the pathophysiology connecting OSA to the renal system is unknown, OSA has been found in up to 60% of patients with end‐stage renal disease and chronic renal failure.34

Patients with pulmonary disorders can be profoundly affected by the normal physiologic changes during sleep, particularly in REM sleep. During REM sleep, all respiratory muscles except the diaphragm become paralyzed. Thus, episodes of marked oxygen desaturation can occur in patients who rely on their accessory muscles for respiration. COPD patients have decreased TST, SWS, and REM sleep. Shortness of breath, nocturnal cough, and wheezing worsen sleep.35 The resulting fatigue and sleep deprivation negatively impact the work of breathing and impair gas exchange. Airflow obstruction tends to worsen in the early morning hours in patients with COPD and asthma, and may be related to the effect of REM on the accessory muscles for respiration. Although used to target CO₂ retention, investigations using bilevel positive airway pressure ventilators (BiPAP) for improving sleep in COPD patients have been limited. Noninvasive positive pressure ventilation (NPPV) appears to acutely improve SE and TST in patients with hypercapnic COPD without significantly improving gas exchange. Other sleep parameters such as sleep architecture and the number of arousals during the night, remain unchanged during NPPV.36

CPAP has several side effects that could worsen sleep, which may explain its poor adherence rate among ambulatory patients.37 Side effects include nasal bridge discomfort, nasal congestion, swallowing air, dry nose, dry or red eyes, noise, ear pain, and rhinitis.38 During hospitalization, efforts should be made to improve patient comfort by resizing ill‐fitting masks, adding heated humidification or nasal steroids to alleviate nasal congestion, or adding a chin strap to reduce air leak and ingestion of air.

Endocrine disorders have also been associated with sleep disruption. Studies suggest that patients with diabetes mellitus have decreased TST and impaired sleep quality due to nocturia and neuropathic pain.39 Inadequate sleep may also affect glucose control. Inadequate quality or quantity of sleep has been shown to be a risk factor for developing Type 2 diabetes mellitus in large prospective studies.40 Sleep duration and quality were significant predictors of increased levels of glycosylated hemoglobin (HbA_1c) in patients with Type 2 diabetes mellitus. Thyroid diseases often coexist with diabetes mellitus. Both hypo‐ and hyperthyroidism have been associated with sleep disruption. Hypothyroidism is associated with daytime somnolence and fatigue. Patients with hypothyroidism tend to have reduced SWS. Hyperthyroid patients often complain of insomnia, which has been attributed to a hypermetabolic state.

Approximately 50% of patients with chronic end‐stage renal disease (ESRD) have insomnia and other sleep disorders.41 Patients often complain of restless leg syndrome (RLS), periodic limb movement disorder (PLMD), bone pain, nausea, and pruritus. The etiology of sleep disorders appears to be related to metabolic derangements associated with ESRD or from coexisting diabetes mellitus.

RLS and PLMD are distinct problems that affect sleep differently. RLS is characterized by an unpleasant crampy, creeping or crawling sensation in the lower extremities that is relieved by movement of the legs.42 RLS symptoms typically occur soon after going to bed, and therefore tend to disrupt sleep onset. The requisite bed rest during hospitalization can worsen RLS, further exacerbating sleep problems.43 Since RLS may partially be caused by disrupted iron metabolism, serum ferritin levels should be evaluated.44 Other conditions associated with RLS include pregnancy, rheumatoid arthritis, fibromyalgia, multiple sclerosis, ESRD, and Parkinson's disease. The differential diagnosis for RLS and PLMD includes neuroleptic‐induced akathisia, peripheral neuropathy, and positional or nocturnal leg cramps. PLMD occurs in about 80% of those with RLS, and is characterized by involuntary limb movements that occur every 20 to 40 seconds during NREM sleep. Unaware of these movements, patients often experience frequent arousals throughout the night, and complain of daytime somnolence and fatigue.42

A pilot study of 35 patients with minimal hepatic encephalopathy found that nearly 50% complained of sleep difficulties.45 Hypothesizing that a dysregulation of histaminergic neurotransmission in cirrhosis alters the sleep‐wake cycle, Spahr et al.46 found that 40% of their patients reported subjective improvement in sleep when administered 25 mg of hydroxyzine, compared to none who received placebo.

Neurologic Disorders

Since the brain and its various neurotransmitter systems are critical in regulating sleep and wakefulness, patients with neurologic disorders have an increased risk of developing sleep disorders. Patients with dementia, other neurodegenerative disorders, epilepsy, and traumatic brain injury (TBI) have a higher prevalence of sleep disturbance and sleep disorders.47 Poststroke patients can develop insomnia or hypersomnia, a reduction in sleep latency, increased sleep, or excessive daytime sleepiness, and are at higher risk for OSA during the first several months after a stroke.48 Specific neurologic lesions may lead to uncommon problems such as inversion of the sleep‐wake cycle, parasomnias, and hallucinatory dream‐like states.

Both Parkinson's disease and Alzheimer's disease are associated with multiple sleep disturbances, which tend to worsen with disease progression.14 Common problems include increased sleep fragmentation and wakefulness, with increases of stage 1 sleep and reductions of SWS and REM. Patients with neurodegenerative disorders also have an increased risk of REM sleep behavior disorder, or RBD.49 RBD is characterized by vivid and unusual dreams, and physically vigorous sleep behaviors that may result in ecchymoses, lacerations, and fractures.50 Fifty percent of patients with TBI reported insomnia symptoms.51 Disorders in initiating and maintaining sleep were the most common complaints among hospitalized patients with TBI. Some patients with TBI may develop circadian rhythm disturbances.52

Pain

A majority of patients with chronic pain, 50% to 70%, complain of impaired sleep.53 Sleep disruption is so common in fibromyalgia (75%) that it is considered to be a key diagnostic symptom.54 In a study investigating the affect of pain on sleep in burn patients, pain was associated with increased intermittent awakenings and prolonged periods of wake time during the night.55 The following day, these patients had poorer pain tolerance and greater pain intensity. Pain causes sleep fragmentation by increasing cortical arousals. Recent evidence suggests that sleep deprivation can increase pain sensitivity by inhibiting opioid protein synthesis or reducing opioid receptor affinity.56

Psychiatric Disorders

Sleep problems are so common in psychiatric conditions that the Diagnostic and Statistical Manual of Mental Disorders (DMS‐IV‐TR) includes sleep disturbance as a diagnostic criterion for a manic episode, and for various depressive, anxiety, and substance abuse disorders.57 The presence of sleep disturbance in hospitalized patients may suggest the presence of an underlying psychiatric disorder that would otherwise go unrecognized. In a survey of 200 general medical patients in a Brazilian hospital, Rocha et al.58 found that 112 (56.5%) complained of insomnia, and 100 (50%) met criteria for at least 1 psychiatric disorder. However, only 3 out of the total number of 200 surveyed (1.5%) were identified as having psychiatric diagnoses in the medical record, and sleep history was not noted in the clinical evaluation. An episode of major depressive disorder was the most common psychiatric diagnosis (35%). In this study, hospitalized patients with insomnia had a 3.6 times higher risk of having major depressive disorder than inpatients without insomnia.

Insomnia has a profound effect on mental health by worsening health‐related quality of life. In a study of outpatients at family medicine, internal medicine, endocrinology, cardiology, and psychiatry clinics in 3 U.S. cities (n = 3,445), insomnia worsened health‐related quality of life nearly as much as CHF or major depressive disorder did.59 Another survey of outpatients found that those with chronic insomnia were nearly 40 times more likely to have major depression and 6 times more likely to have an anxiety disorder compared to those without insomnia.60 Longitudinal studies have found that prior insomnia was associated with 2‐ to 5‐fold increase in the odds of mood and anxiety disorders and suicide.61, 62 Examining prodromes and precursors to mental disorders, Eaton et al.63 found that 47% of those with onset of depression at the 1‐year follow‐up had sleep problems at baseline.

An estimated 65% of patients with major depression have difficulty falling asleep, frequent awakenings, or early morning awakenings.64 Three patterns of sleep architecture abnormalities have been observed in patients with major depression: 1) sleep continuity disturbances characterized by prolonged sleep‐onset, increased wake time during sleep, increased early morning wake time, and decreased TST; 2) decreased proportion and length of SWS; and 3) REM sleep abnormalities such as reduced time to REM sleep, prolonged first REM sleep episode, and increased REM sleep percentage.65 Sleep during a manic episode has been less studied than in depression, but the data suggest that abnormal sleep in mania includes disrupted sleep continuity, shortened REM latency, and increased REM density (REM eye movement activity/total REM sleep time).65

Substance use disorders are also associated with sleep problems. In a survey by Brower et al.66 of patients who were undergoing alcohol rehabilitation, 61% (n = 172) had symptoms of insomnia such as increased sleep latency during the 6 months prior to entering treatment. Approximately 45% of these patients reported using alcohol for the purpose of initiating sleep. Alcohol and illicit substance intoxication and withdrawal are known to be associated with disrupted sleep. However, sleep disturbances may persist long after withdrawal symptoms have abated. Drummond et al. found that some patients continued to have alcohol‐associated sleep problems even after 27 months of abstinence.67 Evidence also suggests that untreated insomnia and other sleep problems may increase the risk of developing substance abuse problems due to self‐medicating with alcohol and other substances to help with sleep.68

Drugs that Affect Sleep

Numerous drugs can alter sleep quantity and quality. Sedatives and opioids may initially help with sleep onset, but impair sleep architecture. Medications used to treat medical and psychiatric illnesses also disrupt sleep (Table 3). The most common agents that impair sleep include antiepileptic drugs, selective serotonin reuptake inhibitors, monoamine oxidase inhibitors, tricyclic antidepressants, antihypertensives, antihistamines, and corticosteroids.

Lipophilic beta antagonists such as propranolol and timolol can increase total wake time, decrease REM sleep, and increase the incidence of nightmares and insomnia.69 Anabolic steroids and beta‐agonist bronchodilator therapy can cause severe anxiety, sleeplessness, and even psychosis. Vasopressor agents such as dopamine can cause cortical activation, leading to increased arousal and reduced SWS.

Hospital Environment

Environmental noise and patient care activities often interfere with sleep in the hospital. They account for about 30% of patient awakenings in ICU patients.70 Noise levels in the ICU have average sound peaks of 150 to 200 dB, and evening peaks >80 dB between midnight and 6 _AM.71 By comparison, the front row seats at a rock concert have sound levels of 110 dB. The high noise level in hospitals has long been implicated as a sleep disruptor,72 but studies in the past decade have found that patient care activities probably contribute more to awakenings than does environmental noise.73 An analysis of critical care nursing routines found that activities such as taking vital signs and giving baths occurred a mean 42.6 times a night per patient.74 Tamburri et al.74 found that patients experienced 2 to 3 hours without interruption on only 6% of the 147 nights studied. Routine daily baths were provided on 55 of the 147 study nights between 2 _AM and 5 _AM, which is unlikely to be an opportune time for most patients.