Lisa Shieh, MD

The ethos of medicine has shifted from paternalistic, physician-driven care to patient autonomy and engagement, in which the physician shares information and advises.^1-3 Although there are ethical, legal, and practical reasons to respect patient preferences,^1-4 patient engagement also fosters quality and safety⁵ and may improve clinical outcomes.^5-8 Patients whose preferences are respected are more likely to trust their doctor, feel empowered, and adhere to treatments.⁹

Providers may partner with patients through shared decision-making (SDM).^10,11 Several SDM models describe the process of providers and patients balancing evidence, preferences and context to arrive at a clinical decision.^12-15 The National Academy of Medicine and the American Academy of Pediatrics has called for more SDM,^16,17 including when clinical evidence is limited,² equally beneficial options exist,¹⁸ clinical stakes are high,¹⁹ and even with deferential patients.²⁰ Despite its value, SDM does not reliably occur^21,22 and SDM training is often unavailable.⁴ Clinical decision tools, patient education aids, and various training interventions have shown promising, although inconsistent results.^{23, 24}

Little is known about SDM in inpatient settings where unique patient, clinician, and environmental factors may influence SDM. This study describes the quality and possible predictors of inpatient SDM during attending rounds in 4 academic training settings. Although SDM may occur anytime during a hospitalization, attending rounds present a valuable opportunity for SDM observation given their centrality to inpatient care and teaching.^25,26 Because attending physicians bear ultimate responsibility for patient management, we examined whether SDM performance varies among attendings within each service. In addition, we tested the hypothesis that service-level, team-level, and patient-level features explain variation in SDM quality more than individual attending physicians. Finally, we compared peer-observer perspectives of SDM behaviors with patient and/or guardian perspectives.

METHODS

Study Design and Setting

This cross-sectional, observational study examined the diversity of SDM practice within and between 4 inpatient services during attending rounds, including the internal medicine and pediatrics services at Stanford University and the University of California, San Francisco (UCSF). Both institutions provide quaternary care to diverse patient populations with approximately half enrolled in Medicare and/or Medicaid.

One institution had 42 internal medicine (Med-1) and 15 pediatric hospitalists (Peds-1) compared to 8 internal medicine (Med-2) and 12 pediatric hospitalists (Peds-2) at the second location. Both pediatric services used family-centered rounds that included discussions between the patients’ families and the whole team. One medicine service used a similar rounding model that did not necessarily involve the patients’ families. In contrast, the smaller medicine service typically began rounds by discussing all patients in a conference room and then visiting select patients afterwards.

From August 2014 to November 2014, peer observers gathered data on team SDM behaviors during attending rounds. After the rounding team departed, nonphysician interviewers surveyed consenting patients’ (or guardians’) views of the SDM experience, yielding paired evaluations for a subset of SDM encounters. Institutional review board approval was obtained from Stanford University and UCSF.

Participants and Inclusion Criteria

Attending physicians were hospitalists who supervised rounds at least 1 month per year, and did not include those conducting the study. All provided verbal assent to be observed on 3 days within a 7-day period. While team composition varied as needed (eg, to include the nurse, pharmacist, interpreter, etc), we restricted study observations to those teams with an attending and at least one learner (eg, resident, intern, medical student) to capture the influence of attending physicians in their training role. Because services vary in number of attendings on staff, rounds assigned per attending, and patients per round, it was not possible to enroll equal sample sizes per service in the study.

 

Nonintensive care unit patients who were deemed medically stable by the team were eligible for peer observation and participation in a subsequent patient interview once during the study period. Pediatric patients were invited for an interview if they were between 13 and 21 years old and had the option of having a parent or guardian present; if the pediatric patients were less than 13 years old or they were not interested in being interviewed, then their parents or guardians were invited to be interviewed. Interpreters were on rounds, and thus, non-English participants were able to participate in the peer observations, but could not participate in patient interviews because interpreters were not available during afternoons for study purposes. Consent was obtained from all participating patients and/or guardians.

Data Collection

Round and Patient Characteristics

Peer observers recorded rounding, team, and patient characteristics using a standardized form. Rounding data included date, attending name, duration of rounds, and patient census. Patient level data included the decision(s) discussed, the seniority of the clinician leading the discussion, team composition, minutes spent discussing the patient (both with the patient and/or guardian and total time), hospitalization week, and patient’s primary language. Additional patient data obtained from electronic health records included age, gender, race, ethnicity, date of admission, and admitting diagnosis.

SDM Measures

Peer-observed SDM behaviors were quantified per patient encounter using the 9-item Rochester Participatory Decision-Making Scale (RPAD), with credit given for SDM behaviors exhibited by anyone on the rounding team (team-level metric).²⁷ Each item was scored on a 3-point scale (0 = absent, 0.5 = partial, and 1 = present) for a maximum of 9 points, with higher scores indicating higher-quality SDM (Peer-RPAD Score). We created semistructured patient interview guides by adapting each RPAD item into layperson language (Patient-RPAD Score) and adding open-ended questions to assess the patient experience.

Peer-Observer Training

Eight peer-observers (7 hospitalists and 1 palliative care physician) were trained to perform RPAD ratings using videos of patient encounters. Initially, raters viewed videos together and discussed ratings for each RPAD item. The observers incorporated behavioral anchors and clinical examples into the development of an RPAD rating guide, which they subsequently used to independently score 4 videos from an online medical communication library.²⁸ These scores were discussed to resolve any differences before 4 additional videos were independently viewed, scored, and compared. Interrater reliability was achieved when the standard deviation of summed SDM scores across raters was less than 1 for all 4 videos.

Patient Interviewers

Interviewers were English-speaking volunteers without formal medical training. They were educated in hospital etiquette by a physician and in administering patient interviews through peer-to-peer role playing and an observation and feedback interview with at least 1 patient.

Data Analysis

The analysis set included every unique patient with whom a medical decision was made by an eligible clinical team. To account for the nested study design (patient-level scores within rounds, rounds within attending, and attendings within service), we used mixed-effects models to estimate mean (summary or item) RPAD score by levels of fixed covariate(s). The models included random effects accounting for attending-level and round-level correlations among scores via variance components, and allowing the attending-level random effect to differ by service. Analyses were performed using SAS version 9.4 (SAS Institute Inc, Cary, NC). We used descriptive statistics to summarize round- and patient-level characteristics.

SDM Variation by Attending and Service

Box plots were used to summarize raw patient-level, Peer-RPAD scores by service and attending. By using the methods described above, we estimated the mean score overall and by service. In both models, we examined the statistical significance of service-specific variation in attending-level random effects by using likelihood-ratio test (LRT) to compare models.

SDM Variation by Round and Patient Characteristics

We used the models described above to identify covariates associated with Peer-RPAD scores. We fit univariate models separately for each covariate, then fit 2 multivariable models, including (1) all covariates and (2) all effects significant in either model at P ≤ .20 according to F tests. For uniformity of presentation, we express continuous covariates categorically; however, we report P values based on continuous versions. Means generated by the multivariable models were calculated at the mean values of all other covariates in model.

Patient-Level RPAD Data

A subsample of patients completed semistructured interviews with analogous RPAD questions. To identify possible selection bias in the full sample, we summarized response rates by service and patient language and modeled Peer-RPAD scores by interview response status. Among responders, we estimated the mean Peer-RPAD and Patient-RPAD scores and their paired differences and correlations, testing for non-zero correlations via the Spearman rank test.

 

RESULTS

All Patient Encounters

A total of 35 attendings (18 medicine, 17 pediatrics) were observed, representing 51% of 69 eligible attendings. By design, study observations included a median of 3 rounds per attending (range 1-5), summing to 88 total rounds (46 medicine, 42 pediatrics) and 783 patient encounters (388 medicine, 395 pediatrics; Table 1).

The median duration of rounding sessions was 1.8 hours, median patient census was 9, and median patient encounter was 13 minutes. The duration of rounds and minutes per patient were longest at Med-2 and shortest at Peds-1. See Table 1 for other team characteristics.

Peer Evaluations of SDM Encounters

Characteristics of Patients

We observed SDM encounters in 254 unique patients (117 medicine, 137 pediatrics), representing 32% of all observed encounters. Patient mean age was 56 years for medicine and 7.4 years for pediatrics. Overall, 54% of patients were white, 11% were Asian, and 10% were African American; race was not reported for 21% of patients. Pediatrics services had more SDM encounters with Hispanic patients (31% vs. 9%) and Spanish-speaking patients (14% vs < 2%; Table 2). Patient complexity ranged from case mix index (CMI) 1.17 (Med-1) to 2. 11 (Peds-1).

Teams spent a median of 13 minutes per SDM encounter, which was not higher than the round median. SDM topics discussed included 47% treatment, 15% diagnostic, 30% both treatment and diagnostic, and 7% other.

Variation in SDM Quality Among Attending Physicians

Overall Peer-RPAD Scores were normally distributed. After adjusting for the nested study design, the overall mean (standard error) score was 4.16 (0.11). Score variability among attendings differed significantly by service (LRT P = .0067). For example, raw scores were lower and more variable among attending physicians at Med-2 than other among attendings in other services (see Appendix Figure in Supporting Information). However, when service was included in the model as a fixed effect, mean scores varied significantly, from 3.0 at Med-2 to 4.7 at Med-1 (P < .0001), but the random variation among attendings no longer differed significantly by service (P = .13). This finding supports the hypothesis that service-level influences are stronger than influences of individual attending physicians, that is, that variation between services exceeded variation among attendings within service.

Aspects of SDM That Are More Prevalent on Rounds

Based on Peer-RPAD item scores, the most frequently observed behaviors across all services included “Matched medical language to the patient’s level of understanding” (Item 6, 0.75) and “Explained the clinical issue or nature of the decision” (Item 1, 0.74; panel A of Figure). The least frequently observed behaviors included “Asked if patient had any questions” (Item 7, 0.34), “Examined barriers to follow-through with the treatment plan” (Item 4, 0.15), and “Checked understanding of the patient’s point of view” (Item 9, 0.06).

Rounds and Patient Characteristics Associated With Peer-RPAD Scores

In univariate models, Peer-RPAD scores decreased significantly with round-level average minutes per patient and were elevated during a patient’s second week of hospitalization. In the multivariable model including all covariates in Table 3, mean Peer-RPAD scores varied by service (lower at Med-2 than elsewhere), patient gender (slightly higher among women and girls), week of hospitalization (highest during the second week), and time spent with the patient and/or guardian (more time correlated with higher scores). In a reduced multivariable model restricted to the covariates that were statistically significant in either model (P ≤ .20), all 5 associations remained significant P ≤ .05. However, the difference in means by gender was only 0.3, and only 18% of patients were hospitalized for more than 1 week.

Patient-RPAD Results: Dissimilar Perspectives of Patients and/or Guardians and Physician Observers

Of 254 peer-evaluated SDM encounters, 149 (59%) patients and/or guardians were available and consented to same-day interviews, allowing comparison of paired peer and patient evaluations of SDM in this subset. The response rate was 66% among patients whose primary language was English versus 15% among others. Peer-RPAD scores by interview response status were similar overall (responders, 4.17; nonresponders, 4.13; P = .83) and by service (interaction P = .30).

Among responders, mean Patient-RPAD scores were 6.8 to 7.1 for medicine services and 7.6 to 7.8 for pediatric services (P = .01). The overall mean Patient-RPAD score, 7.46, was significantly greater than the paired Peer-RPAD score by 3.5 (P = .011); however, correlations were not statistically significantly different from 0 (by service, each P > .12).

To understand drivers of the differences between Peer-RPAD and Patient-RPAD scores, we analyzed findings by item. Each mean patient-item score exceeded its peer counterpart (P ≤ .01; panel B of Figure). Peer-item scores fell below 33% on 2 items (Items 9 and 4) and only exceeded 67% on 2 items (Items 1 and 6), whereas patient-item scores ranged from 60% (Item 8) to 97% (Item 7). Three paired differences exceeded 50% (Items 9, 4, and 7) and 3 were below 20% (Items 6, 8 and 1), underlying the lack of correlation between peer and patient scores.

 

DISCUSSION

In this multisite study of SDM during inpatient attending rounds, SDM quality, specific SDM behaviors, and factors contributing to SDM were identified. Our study found an adjusted overall Peer-RPAD Score of 4.4 out of 9, and found the following 3 SDM elements most needing improvement according to trained peer observers: (1) “Checking understanding of the patient’s perspective”, (2) “Examining barriers to follow-through with the treatment plan”, and (3) “Asking if the patient has questions.” Areas of strength included explaining the clinical issue or nature of the decision and matching medical language to the patient’s level of understanding, with each rated highly by both peer-observers and patients. Broadly speaking, physicians were skillful in delivering information to patients but failed to solicit input from patients. Characteristics associated with increased SDM in the multivariate analysis included the following: service, patient gender, timing of rounds during patient’s hospital stay, and amount of time rounding with each patient.

Patients similarly found that physicians could improve their abilities to elicit information from patients and families, noting the 3 lowest patient-rated SDM elements were as follows: (1) asking open-ended questions, (2) discussing alternatives or uncertainties, and (3) discussing barriers to treatment plan follow through. Overall, patients and guardians perceived the quantity and quality of SDM on rounds more favorably than peer observers, which is consistent with other studies of patient perceptions of communication. ^29-31 It is possible that patient ratings are more influenced by demand characteristics, fear of negatively impacting their patient-provider relationships, and conflation of overall satisfaction with quality of communication.³² This difference in patient perception of SDM is worthy of further study.

Prior work has revealed that SDM may occur infrequently during inpatient rounds.¹¹ This study further elucidates specific SDM behaviors used along with univariate and multivariate modeling to explore possible contributing factors. The strengths and weaknesses found were similar at all 4 services and the influence of the service was more important than variability across attendings. This study’s findings are similar to a study by Shields et al.,³³ in which the findings in a geographically different outpatient setting 10 years earlier suggesting global and enduring challenges to SDM. To our knowledge, this is the first published study to characterize inpatient SDM behaviors and may serve as the basis for future interventions.

Although the item-level components were ranked similarly across services, on average the summary Peer-RPAD score was lowest at Med-2, where we observed high variability within and between attendings, and was highest at Med-1, where variability was low. Med-2 carried the highest caseload and held the longest rounds, while Med-1 carried the lowest caseload, suggesting that modifiable burdens may hamper SDM performance. Prior studies suggest that patients are often selected based on teaching opportunities, immediate medical need and being newly admitted.³⁴ The high scores at Med-1 may reflect that service’s prediscussion of patients during card-flipping rounds or their selection of which patients to round on as a team. Consistent with prior studies^29,35 of SDM and the family-centered rounding model, which includes the involvement of nurses, respiratory therapists, pharmacists, case managers, social workers, and interpreters on rounds, both pediatrics services showed higher SDM scores.

In contrast to prior studies,^34,36 team size and number of learners did not affect SDM performance, nor did decision type. Despite teams having up to 17 members, 8 learners, and 14 complex patients, SDM scores did not vary significantly by team. Nonetheless, trends were in the directions expected: Scores tended to decrease as the team size or the percentage of trainees grew, and increased with the seniority of the presenting physician. Interestingly, SDM performance decreased with round-average minutes per patient, which may be measuring on-going intensity across cases that leads to exhaustion. Statistically significant patient factors for increased SDM included longer duration of patient encounters, second week of hospital stay, and female patient gender. Although we anticipated that the high number of decisions made early in hospitalization would facilitate higher SDM scores, continuity and stronger patient-provider relationships may enhance SDM.³⁶ We report service-specific team and patient characteristics, in addition to SDM findings in anticipation that some readers will identify with 1 service more than others.

This study has several important limitations. First, our peer observers were not blinded and primarily observed encounters at their own site. To minimize bias, observers periodically rated videos to recalibrate RPAD scoring. Second, additional SDM conversations with a patient and/or guardian may have occurred outside of rounds and were not captured, and poor patient recall may have affected Patient-RPAD scores despite interviewer prompts and timeliness of interviews within 12 hours of rounds. Third, there might have been a selection bias for the one service who selected a smaller number of patients to see, compared with the three other services that performed bedside rounds on all patients. It is possible that attending physicians selected patients who were deemed most able to have SDM conversations, thus affecting RPAD scores on that service. Fourth, study services had fewer patients on average than other academic hospitals (median 9, range 3-14), which might limit its generalizability. Last, as in any observational study, there is always the possibility of the Hawthorne effect. However, neither teams nor patients knew the study objectives.

Nevertheless, important findings emerged through the use of RPAD Scores to evaluate inpatient SDM practices. In particular, we found that to increase SDM quality in inpatient settings, practitioners should (1) check their understanding of the patient’s perspective, (2) examine barriers to follow-through with the treatment plan, and (3) ask if the patient has questions. Variation among services remained very influential after adjusting for team and patient characteristics, which suggests that “climate” or service culture should be targeted by an intervention, rather than individual attendings or subgroups defined by team or patient characteristics. Notably, team size, number of learners, patient census, and type of decision being made did not affect SDM performance, suggesting that even large, busy services can perform SDM if properly trained.

 

Acknowledgments

The authors thank the patients, families, pediatric and internal medicine residents, and hospitalists at Stanford School of Medicine and University of California, San Francisco School of Medicine for their participation in this study. We would also like to thank the student volunteers who collected patient perspectives on the encounters.

Disclosure 

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by an NIH/NCCIH grant R25 AT006573.

The ethos of medicine has shifted from paternalistic, physician-driven care to patient autonomy and engagement, in which the physician shares information and advises.^1-3 Although there are ethical, legal, and practical reasons to respect patient preferences,^1-4 patient engagement also fosters quality and safety⁵ and may improve clinical outcomes.^5-8 Patients whose preferences are respected are more likely to trust their doctor, feel empowered, and adhere to treatments.⁹

Providers may partner with patients through shared decision-making (SDM).^10,11 Several SDM models describe the process of providers and patients balancing evidence, preferences and context to arrive at a clinical decision.^12-15 The National Academy of Medicine and the American Academy of Pediatrics has called for more SDM,^16,17 including when clinical evidence is limited,² equally beneficial options exist,¹⁸ clinical stakes are high,¹⁹ and even with deferential patients.²⁰ Despite its value, SDM does not reliably occur^21,22 and SDM training is often unavailable.⁴ Clinical decision tools, patient education aids, and various training interventions have shown promising, although inconsistent results.^{23, 24}

Little is known about SDM in inpatient settings where unique patient, clinician, and environmental factors may influence SDM. This study describes the quality and possible predictors of inpatient SDM during attending rounds in 4 academic training settings. Although SDM may occur anytime during a hospitalization, attending rounds present a valuable opportunity for SDM observation given their centrality to inpatient care and teaching.^25,26 Because attending physicians bear ultimate responsibility for patient management, we examined whether SDM performance varies among attendings within each service. In addition, we tested the hypothesis that service-level, team-level, and patient-level features explain variation in SDM quality more than individual attending physicians. Finally, we compared peer-observer perspectives of SDM behaviors with patient and/or guardian perspectives.

METHODS

Study Design and Setting

This cross-sectional, observational study examined the diversity of SDM practice within and between 4 inpatient services during attending rounds, including the internal medicine and pediatrics services at Stanford University and the University of California, San Francisco (UCSF). Both institutions provide quaternary care to diverse patient populations with approximately half enrolled in Medicare and/or Medicaid.

One institution had 42 internal medicine (Med-1) and 15 pediatric hospitalists (Peds-1) compared to 8 internal medicine (Med-2) and 12 pediatric hospitalists (Peds-2) at the second location. Both pediatric services used family-centered rounds that included discussions between the patients’ families and the whole team. One medicine service used a similar rounding model that did not necessarily involve the patients’ families. In contrast, the smaller medicine service typically began rounds by discussing all patients in a conference room and then visiting select patients afterwards.

From August 2014 to November 2014, peer observers gathered data on team SDM behaviors during attending rounds. After the rounding team departed, nonphysician interviewers surveyed consenting patients’ (or guardians’) views of the SDM experience, yielding paired evaluations for a subset of SDM encounters. Institutional review board approval was obtained from Stanford University and UCSF.

Participants and Inclusion Criteria

Attending physicians were hospitalists who supervised rounds at least 1 month per year, and did not include those conducting the study. All provided verbal assent to be observed on 3 days within a 7-day period. While team composition varied as needed (eg, to include the nurse, pharmacist, interpreter, etc), we restricted study observations to those teams with an attending and at least one learner (eg, resident, intern, medical student) to capture the influence of attending physicians in their training role. Because services vary in number of attendings on staff, rounds assigned per attending, and patients per round, it was not possible to enroll equal sample sizes per service in the study.

 

Nonintensive care unit patients who were deemed medically stable by the team were eligible for peer observation and participation in a subsequent patient interview once during the study period. Pediatric patients were invited for an interview if they were between 13 and 21 years old and had the option of having a parent or guardian present; if the pediatric patients were less than 13 years old or they were not interested in being interviewed, then their parents or guardians were invited to be interviewed. Interpreters were on rounds, and thus, non-English participants were able to participate in the peer observations, but could not participate in patient interviews because interpreters were not available during afternoons for study purposes. Consent was obtained from all participating patients and/or guardians.

Data Collection

Round and Patient Characteristics

Peer observers recorded rounding, team, and patient characteristics using a standardized form. Rounding data included date, attending name, duration of rounds, and patient census. Patient level data included the decision(s) discussed, the seniority of the clinician leading the discussion, team composition, minutes spent discussing the patient (both with the patient and/or guardian and total time), hospitalization week, and patient’s primary language. Additional patient data obtained from electronic health records included age, gender, race, ethnicity, date of admission, and admitting diagnosis.

SDM Measures

Peer-observed SDM behaviors were quantified per patient encounter using the 9-item Rochester Participatory Decision-Making Scale (RPAD), with credit given for SDM behaviors exhibited by anyone on the rounding team (team-level metric).²⁷ Each item was scored on a 3-point scale (0 = absent, 0.5 = partial, and 1 = present) for a maximum of 9 points, with higher scores indicating higher-quality SDM (Peer-RPAD Score). We created semistructured patient interview guides by adapting each RPAD item into layperson language (Patient-RPAD Score) and adding open-ended questions to assess the patient experience.

Peer-Observer Training

Eight peer-observers (7 hospitalists and 1 palliative care physician) were trained to perform RPAD ratings using videos of patient encounters. Initially, raters viewed videos together and discussed ratings for each RPAD item. The observers incorporated behavioral anchors and clinical examples into the development of an RPAD rating guide, which they subsequently used to independently score 4 videos from an online medical communication library.²⁸ These scores were discussed to resolve any differences before 4 additional videos were independently viewed, scored, and compared. Interrater reliability was achieved when the standard deviation of summed SDM scores across raters was less than 1 for all 4 videos.

Patient Interviewers

Interviewers were English-speaking volunteers without formal medical training. They were educated in hospital etiquette by a physician and in administering patient interviews through peer-to-peer role playing and an observation and feedback interview with at least 1 patient.

Data Analysis

The analysis set included every unique patient with whom a medical decision was made by an eligible clinical team. To account for the nested study design (patient-level scores within rounds, rounds within attending, and attendings within service), we used mixed-effects models to estimate mean (summary or item) RPAD score by levels of fixed covariate(s). The models included random effects accounting for attending-level and round-level correlations among scores via variance components, and allowing the attending-level random effect to differ by service. Analyses were performed using SAS version 9.4 (SAS Institute Inc, Cary, NC). We used descriptive statistics to summarize round- and patient-level characteristics.

SDM Variation by Attending and Service

Box plots were used to summarize raw patient-level, Peer-RPAD scores by service and attending. By using the methods described above, we estimated the mean score overall and by service. In both models, we examined the statistical significance of service-specific variation in attending-level random effects by using likelihood-ratio test (LRT) to compare models.

SDM Variation by Round and Patient Characteristics

We used the models described above to identify covariates associated with Peer-RPAD scores. We fit univariate models separately for each covariate, then fit 2 multivariable models, including (1) all covariates and (2) all effects significant in either model at P ≤ .20 according to F tests. For uniformity of presentation, we express continuous covariates categorically; however, we report P values based on continuous versions. Means generated by the multivariable models were calculated at the mean values of all other covariates in model.

Patient-Level RPAD Data

A subsample of patients completed semistructured interviews with analogous RPAD questions. To identify possible selection bias in the full sample, we summarized response rates by service and patient language and modeled Peer-RPAD scores by interview response status. Among responders, we estimated the mean Peer-RPAD and Patient-RPAD scores and their paired differences and correlations, testing for non-zero correlations via the Spearman rank test.

 

RESULTS

All Patient Encounters

A total of 35 attendings (18 medicine, 17 pediatrics) were observed, representing 51% of 69 eligible attendings. By design, study observations included a median of 3 rounds per attending (range 1-5), summing to 88 total rounds (46 medicine, 42 pediatrics) and 783 patient encounters (388 medicine, 395 pediatrics; Table 1).

The median duration of rounding sessions was 1.8 hours, median patient census was 9, and median patient encounter was 13 minutes. The duration of rounds and minutes per patient were longest at Med-2 and shortest at Peds-1. See Table 1 for other team characteristics.

Peer Evaluations of SDM Encounters

Characteristics of Patients

We observed SDM encounters in 254 unique patients (117 medicine, 137 pediatrics), representing 32% of all observed encounters. Patient mean age was 56 years for medicine and 7.4 years for pediatrics. Overall, 54% of patients were white, 11% were Asian, and 10% were African American; race was not reported for 21% of patients. Pediatrics services had more SDM encounters with Hispanic patients (31% vs. 9%) and Spanish-speaking patients (14% vs < 2%; Table 2). Patient complexity ranged from case mix index (CMI) 1.17 (Med-1) to 2. 11 (Peds-1).

Teams spent a median of 13 minutes per SDM encounter, which was not higher than the round median. SDM topics discussed included 47% treatment, 15% diagnostic, 30% both treatment and diagnostic, and 7% other.

Variation in SDM Quality Among Attending Physicians

Overall Peer-RPAD Scores were normally distributed. After adjusting for the nested study design, the overall mean (standard error) score was 4.16 (0.11). Score variability among attendings differed significantly by service (LRT P = .0067). For example, raw scores were lower and more variable among attending physicians at Med-2 than other among attendings in other services (see Appendix Figure in Supporting Information). However, when service was included in the model as a fixed effect, mean scores varied significantly, from 3.0 at Med-2 to 4.7 at Med-1 (P < .0001), but the random variation among attendings no longer differed significantly by service (P = .13). This finding supports the hypothesis that service-level influences are stronger than influences of individual attending physicians, that is, that variation between services exceeded variation among attendings within service.

Aspects of SDM That Are More Prevalent on Rounds

Based on Peer-RPAD item scores, the most frequently observed behaviors across all services included “Matched medical language to the patient’s level of understanding” (Item 6, 0.75) and “Explained the clinical issue or nature of the decision” (Item 1, 0.74; panel A of Figure). The least frequently observed behaviors included “Asked if patient had any questions” (Item 7, 0.34), “Examined barriers to follow-through with the treatment plan” (Item 4, 0.15), and “Checked understanding of the patient’s point of view” (Item 9, 0.06).

Rounds and Patient Characteristics Associated With Peer-RPAD Scores

In univariate models, Peer-RPAD scores decreased significantly with round-level average minutes per patient and were elevated during a patient’s second week of hospitalization. In the multivariable model including all covariates in Table 3, mean Peer-RPAD scores varied by service (lower at Med-2 than elsewhere), patient gender (slightly higher among women and girls), week of hospitalization (highest during the second week), and time spent with the patient and/or guardian (more time correlated with higher scores). In a reduced multivariable model restricted to the covariates that were statistically significant in either model (P ≤ .20), all 5 associations remained significant P ≤ .05. However, the difference in means by gender was only 0.3, and only 18% of patients were hospitalized for more than 1 week.

Patient-RPAD Results: Dissimilar Perspectives of Patients and/or Guardians and Physician Observers

Of 254 peer-evaluated SDM encounters, 149 (59%) patients and/or guardians were available and consented to same-day interviews, allowing comparison of paired peer and patient evaluations of SDM in this subset. The response rate was 66% among patients whose primary language was English versus 15% among others. Peer-RPAD scores by interview response status were similar overall (responders, 4.17; nonresponders, 4.13; P = .83) and by service (interaction P = .30).

Among responders, mean Patient-RPAD scores were 6.8 to 7.1 for medicine services and 7.6 to 7.8 for pediatric services (P = .01). The overall mean Patient-RPAD score, 7.46, was significantly greater than the paired Peer-RPAD score by 3.5 (P = .011); however, correlations were not statistically significantly different from 0 (by service, each P > .12).

To understand drivers of the differences between Peer-RPAD and Patient-RPAD scores, we analyzed findings by item. Each mean patient-item score exceeded its peer counterpart (P ≤ .01; panel B of Figure). Peer-item scores fell below 33% on 2 items (Items 9 and 4) and only exceeded 67% on 2 items (Items 1 and 6), whereas patient-item scores ranged from 60% (Item 8) to 97% (Item 7). Three paired differences exceeded 50% (Items 9, 4, and 7) and 3 were below 20% (Items 6, 8 and 1), underlying the lack of correlation between peer and patient scores.

 

DISCUSSION

In this multisite study of SDM during inpatient attending rounds, SDM quality, specific SDM behaviors, and factors contributing to SDM were identified. Our study found an adjusted overall Peer-RPAD Score of 4.4 out of 9, and found the following 3 SDM elements most needing improvement according to trained peer observers: (1) “Checking understanding of the patient’s perspective”, (2) “Examining barriers to follow-through with the treatment plan”, and (3) “Asking if the patient has questions.” Areas of strength included explaining the clinical issue or nature of the decision and matching medical language to the patient’s level of understanding, with each rated highly by both peer-observers and patients. Broadly speaking, physicians were skillful in delivering information to patients but failed to solicit input from patients. Characteristics associated with increased SDM in the multivariate analysis included the following: service, patient gender, timing of rounds during patient’s hospital stay, and amount of time rounding with each patient.

Patients similarly found that physicians could improve their abilities to elicit information from patients and families, noting the 3 lowest patient-rated SDM elements were as follows: (1) asking open-ended questions, (2) discussing alternatives or uncertainties, and (3) discussing barriers to treatment plan follow through. Overall, patients and guardians perceived the quantity and quality of SDM on rounds more favorably than peer observers, which is consistent with other studies of patient perceptions of communication. ^29-31 It is possible that patient ratings are more influenced by demand characteristics, fear of negatively impacting their patient-provider relationships, and conflation of overall satisfaction with quality of communication.³² This difference in patient perception of SDM is worthy of further study.

Prior work has revealed that SDM may occur infrequently during inpatient rounds.¹¹ This study further elucidates specific SDM behaviors used along with univariate and multivariate modeling to explore possible contributing factors. The strengths and weaknesses found were similar at all 4 services and the influence of the service was more important than variability across attendings. This study’s findings are similar to a study by Shields et al.,³³ in which the findings in a geographically different outpatient setting 10 years earlier suggesting global and enduring challenges to SDM. To our knowledge, this is the first published study to characterize inpatient SDM behaviors and may serve as the basis for future interventions.

Although the item-level components were ranked similarly across services, on average the summary Peer-RPAD score was lowest at Med-2, where we observed high variability within and between attendings, and was highest at Med-1, where variability was low. Med-2 carried the highest caseload and held the longest rounds, while Med-1 carried the lowest caseload, suggesting that modifiable burdens may hamper SDM performance. Prior studies suggest that patients are often selected based on teaching opportunities, immediate medical need and being newly admitted.³⁴ The high scores at Med-1 may reflect that service’s prediscussion of patients during card-flipping rounds or their selection of which patients to round on as a team. Consistent with prior studies^29,35 of SDM and the family-centered rounding model, which includes the involvement of nurses, respiratory therapists, pharmacists, case managers, social workers, and interpreters on rounds, both pediatrics services showed higher SDM scores.

In contrast to prior studies,^34,36 team size and number of learners did not affect SDM performance, nor did decision type. Despite teams having up to 17 members, 8 learners, and 14 complex patients, SDM scores did not vary significantly by team. Nonetheless, trends were in the directions expected: Scores tended to decrease as the team size or the percentage of trainees grew, and increased with the seniority of the presenting physician. Interestingly, SDM performance decreased with round-average minutes per patient, which may be measuring on-going intensity across cases that leads to exhaustion. Statistically significant patient factors for increased SDM included longer duration of patient encounters, second week of hospital stay, and female patient gender. Although we anticipated that the high number of decisions made early in hospitalization would facilitate higher SDM scores, continuity and stronger patient-provider relationships may enhance SDM.³⁶ We report service-specific team and patient characteristics, in addition to SDM findings in anticipation that some readers will identify with 1 service more than others.

This study has several important limitations. First, our peer observers were not blinded and primarily observed encounters at their own site. To minimize bias, observers periodically rated videos to recalibrate RPAD scoring. Second, additional SDM conversations with a patient and/or guardian may have occurred outside of rounds and were not captured, and poor patient recall may have affected Patient-RPAD scores despite interviewer prompts and timeliness of interviews within 12 hours of rounds. Third, there might have been a selection bias for the one service who selected a smaller number of patients to see, compared with the three other services that performed bedside rounds on all patients. It is possible that attending physicians selected patients who were deemed most able to have SDM conversations, thus affecting RPAD scores on that service. Fourth, study services had fewer patients on average than other academic hospitals (median 9, range 3-14), which might limit its generalizability. Last, as in any observational study, there is always the possibility of the Hawthorne effect. However, neither teams nor patients knew the study objectives.

Nevertheless, important findings emerged through the use of RPAD Scores to evaluate inpatient SDM practices. In particular, we found that to increase SDM quality in inpatient settings, practitioners should (1) check their understanding of the patient’s perspective, (2) examine barriers to follow-through with the treatment plan, and (3) ask if the patient has questions. Variation among services remained very influential after adjusting for team and patient characteristics, which suggests that “climate” or service culture should be targeted by an intervention, rather than individual attendings or subgroups defined by team or patient characteristics. Notably, team size, number of learners, patient census, and type of decision being made did not affect SDM performance, suggesting that even large, busy services can perform SDM if properly trained.

 

Acknowledgments

The authors thank the patients, families, pediatric and internal medicine residents, and hospitalists at Stanford School of Medicine and University of California, San Francisco School of Medicine for their participation in this study. We would also like to thank the student volunteers who collected patient perspectives on the encounters.

Disclosure 

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by an NIH/NCCIH grant R25 AT006573.

The ethos of medicine has shifted from paternalistic, physician-driven care to patient autonomy and engagement, in which the physician shares information and advises.^1-3 Although there are ethical, legal, and practical reasons to respect patient preferences,^1-4 patient engagement also fosters quality and safety⁵ and may improve clinical outcomes.^5-8 Patients whose preferences are respected are more likely to trust their doctor, feel empowered, and adhere to treatments.⁹

Providers may partner with patients through shared decision-making (SDM).^10,11 Several SDM models describe the process of providers and patients balancing evidence, preferences and context to arrive at a clinical decision.^12-15 The National Academy of Medicine and the American Academy of Pediatrics has called for more SDM,^16,17 including when clinical evidence is limited,² equally beneficial options exist,¹⁸ clinical stakes are high,¹⁹ and even with deferential patients.²⁰ Despite its value, SDM does not reliably occur^21,22 and SDM training is often unavailable.⁴ Clinical decision tools, patient education aids, and various training interventions have shown promising, although inconsistent results.^{23, 24}

Little is known about SDM in inpatient settings where unique patient, clinician, and environmental factors may influence SDM. This study describes the quality and possible predictors of inpatient SDM during attending rounds in 4 academic training settings. Although SDM may occur anytime during a hospitalization, attending rounds present a valuable opportunity for SDM observation given their centrality to inpatient care and teaching.^25,26 Because attending physicians bear ultimate responsibility for patient management, we examined whether SDM performance varies among attendings within each service. In addition, we tested the hypothesis that service-level, team-level, and patient-level features explain variation in SDM quality more than individual attending physicians. Finally, we compared peer-observer perspectives of SDM behaviors with patient and/or guardian perspectives.

METHODS

Study Design and Setting

This cross-sectional, observational study examined the diversity of SDM practice within and between 4 inpatient services during attending rounds, including the internal medicine and pediatrics services at Stanford University and the University of California, San Francisco (UCSF). Both institutions provide quaternary care to diverse patient populations with approximately half enrolled in Medicare and/or Medicaid.

One institution had 42 internal medicine (Med-1) and 15 pediatric hospitalists (Peds-1) compared to 8 internal medicine (Med-2) and 12 pediatric hospitalists (Peds-2) at the second location. Both pediatric services used family-centered rounds that included discussions between the patients’ families and the whole team. One medicine service used a similar rounding model that did not necessarily involve the patients’ families. In contrast, the smaller medicine service typically began rounds by discussing all patients in a conference room and then visiting select patients afterwards.

From August 2014 to November 2014, peer observers gathered data on team SDM behaviors during attending rounds. After the rounding team departed, nonphysician interviewers surveyed consenting patients’ (or guardians’) views of the SDM experience, yielding paired evaluations for a subset of SDM encounters. Institutional review board approval was obtained from Stanford University and UCSF.

Participants and Inclusion Criteria

Attending physicians were hospitalists who supervised rounds at least 1 month per year, and did not include those conducting the study. All provided verbal assent to be observed on 3 days within a 7-day period. While team composition varied as needed (eg, to include the nurse, pharmacist, interpreter, etc), we restricted study observations to those teams with an attending and at least one learner (eg, resident, intern, medical student) to capture the influence of attending physicians in their training role. Because services vary in number of attendings on staff, rounds assigned per attending, and patients per round, it was not possible to enroll equal sample sizes per service in the study.

 

Nonintensive care unit patients who were deemed medically stable by the team were eligible for peer observation and participation in a subsequent patient interview once during the study period. Pediatric patients were invited for an interview if they were between 13 and 21 years old and had the option of having a parent or guardian present; if the pediatric patients were less than 13 years old or they were not interested in being interviewed, then their parents or guardians were invited to be interviewed. Interpreters were on rounds, and thus, non-English participants were able to participate in the peer observations, but could not participate in patient interviews because interpreters were not available during afternoons for study purposes. Consent was obtained from all participating patients and/or guardians.

Data Collection

Round and Patient Characteristics

Peer observers recorded rounding, team, and patient characteristics using a standardized form. Rounding data included date, attending name, duration of rounds, and patient census. Patient level data included the decision(s) discussed, the seniority of the clinician leading the discussion, team composition, minutes spent discussing the patient (both with the patient and/or guardian and total time), hospitalization week, and patient’s primary language. Additional patient data obtained from electronic health records included age, gender, race, ethnicity, date of admission, and admitting diagnosis.

SDM Measures

Peer-observed SDM behaviors were quantified per patient encounter using the 9-item Rochester Participatory Decision-Making Scale (RPAD), with credit given for SDM behaviors exhibited by anyone on the rounding team (team-level metric).²⁷ Each item was scored on a 3-point scale (0 = absent, 0.5 = partial, and 1 = present) for a maximum of 9 points, with higher scores indicating higher-quality SDM (Peer-RPAD Score). We created semistructured patient interview guides by adapting each RPAD item into layperson language (Patient-RPAD Score) and adding open-ended questions to assess the patient experience.

Peer-Observer Training

Eight peer-observers (7 hospitalists and 1 palliative care physician) were trained to perform RPAD ratings using videos of patient encounters. Initially, raters viewed videos together and discussed ratings for each RPAD item. The observers incorporated behavioral anchors and clinical examples into the development of an RPAD rating guide, which they subsequently used to independently score 4 videos from an online medical communication library.²⁸ These scores were discussed to resolve any differences before 4 additional videos were independently viewed, scored, and compared. Interrater reliability was achieved when the standard deviation of summed SDM scores across raters was less than 1 for all 4 videos.

Patient Interviewers

Interviewers were English-speaking volunteers without formal medical training. They were educated in hospital etiquette by a physician and in administering patient interviews through peer-to-peer role playing and an observation and feedback interview with at least 1 patient.

Data Analysis

The analysis set included every unique patient with whom a medical decision was made by an eligible clinical team. To account for the nested study design (patient-level scores within rounds, rounds within attending, and attendings within service), we used mixed-effects models to estimate mean (summary or item) RPAD score by levels of fixed covariate(s). The models included random effects accounting for attending-level and round-level correlations among scores via variance components, and allowing the attending-level random effect to differ by service. Analyses were performed using SAS version 9.4 (SAS Institute Inc, Cary, NC). We used descriptive statistics to summarize round- and patient-level characteristics.

SDM Variation by Attending and Service

Box plots were used to summarize raw patient-level, Peer-RPAD scores by service and attending. By using the methods described above, we estimated the mean score overall and by service. In both models, we examined the statistical significance of service-specific variation in attending-level random effects by using likelihood-ratio test (LRT) to compare models.

SDM Variation by Round and Patient Characteristics

We used the models described above to identify covariates associated with Peer-RPAD scores. We fit univariate models separately for each covariate, then fit 2 multivariable models, including (1) all covariates and (2) all effects significant in either model at P ≤ .20 according to F tests. For uniformity of presentation, we express continuous covariates categorically; however, we report P values based on continuous versions. Means generated by the multivariable models were calculated at the mean values of all other covariates in model.

Patient-Level RPAD Data

A subsample of patients completed semistructured interviews with analogous RPAD questions. To identify possible selection bias in the full sample, we summarized response rates by service and patient language and modeled Peer-RPAD scores by interview response status. Among responders, we estimated the mean Peer-RPAD and Patient-RPAD scores and their paired differences and correlations, testing for non-zero correlations via the Spearman rank test.

 

RESULTS

All Patient Encounters

A total of 35 attendings (18 medicine, 17 pediatrics) were observed, representing 51% of 69 eligible attendings. By design, study observations included a median of 3 rounds per attending (range 1-5), summing to 88 total rounds (46 medicine, 42 pediatrics) and 783 patient encounters (388 medicine, 395 pediatrics; Table 1).

The median duration of rounding sessions was 1.8 hours, median patient census was 9, and median patient encounter was 13 minutes. The duration of rounds and minutes per patient were longest at Med-2 and shortest at Peds-1. See Table 1 for other team characteristics.

Peer Evaluations of SDM Encounters

Characteristics of Patients

We observed SDM encounters in 254 unique patients (117 medicine, 137 pediatrics), representing 32% of all observed encounters. Patient mean age was 56 years for medicine and 7.4 years for pediatrics. Overall, 54% of patients were white, 11% were Asian, and 10% were African American; race was not reported for 21% of patients. Pediatrics services had more SDM encounters with Hispanic patients (31% vs. 9%) and Spanish-speaking patients (14% vs < 2%; Table 2). Patient complexity ranged from case mix index (CMI) 1.17 (Med-1) to 2. 11 (Peds-1).

Teams spent a median of 13 minutes per SDM encounter, which was not higher than the round median. SDM topics discussed included 47% treatment, 15% diagnostic, 30% both treatment and diagnostic, and 7% other.

Variation in SDM Quality Among Attending Physicians

Overall Peer-RPAD Scores were normally distributed. After adjusting for the nested study design, the overall mean (standard error) score was 4.16 (0.11). Score variability among attendings differed significantly by service (LRT P = .0067). For example, raw scores were lower and more variable among attending physicians at Med-2 than other among attendings in other services (see Appendix Figure in Supporting Information). However, when service was included in the model as a fixed effect, mean scores varied significantly, from 3.0 at Med-2 to 4.7 at Med-1 (P < .0001), but the random variation among attendings no longer differed significantly by service (P = .13). This finding supports the hypothesis that service-level influences are stronger than influences of individual attending physicians, that is, that variation between services exceeded variation among attendings within service.

Aspects of SDM That Are More Prevalent on Rounds

Based on Peer-RPAD item scores, the most frequently observed behaviors across all services included “Matched medical language to the patient’s level of understanding” (Item 6, 0.75) and “Explained the clinical issue or nature of the decision” (Item 1, 0.74; panel A of Figure). The least frequently observed behaviors included “Asked if patient had any questions” (Item 7, 0.34), “Examined barriers to follow-through with the treatment plan” (Item 4, 0.15), and “Checked understanding of the patient’s point of view” (Item 9, 0.06).

Rounds and Patient Characteristics Associated With Peer-RPAD Scores

In univariate models, Peer-RPAD scores decreased significantly with round-level average minutes per patient and were elevated during a patient’s second week of hospitalization. In the multivariable model including all covariates in Table 3, mean Peer-RPAD scores varied by service (lower at Med-2 than elsewhere), patient gender (slightly higher among women and girls), week of hospitalization (highest during the second week), and time spent with the patient and/or guardian (more time correlated with higher scores). In a reduced multivariable model restricted to the covariates that were statistically significant in either model (P ≤ .20), all 5 associations remained significant P ≤ .05. However, the difference in means by gender was only 0.3, and only 18% of patients were hospitalized for more than 1 week.

Patient-RPAD Results: Dissimilar Perspectives of Patients and/or Guardians and Physician Observers

Of 254 peer-evaluated SDM encounters, 149 (59%) patients and/or guardians were available and consented to same-day interviews, allowing comparison of paired peer and patient evaluations of SDM in this subset. The response rate was 66% among patients whose primary language was English versus 15% among others. Peer-RPAD scores by interview response status were similar overall (responders, 4.17; nonresponders, 4.13; P = .83) and by service (interaction P = .30).

Among responders, mean Patient-RPAD scores were 6.8 to 7.1 for medicine services and 7.6 to 7.8 for pediatric services (P = .01). The overall mean Patient-RPAD score, 7.46, was significantly greater than the paired Peer-RPAD score by 3.5 (P = .011); however, correlations were not statistically significantly different from 0 (by service, each P > .12).

To understand drivers of the differences between Peer-RPAD and Patient-RPAD scores, we analyzed findings by item. Each mean patient-item score exceeded its peer counterpart (P ≤ .01; panel B of Figure). Peer-item scores fell below 33% on 2 items (Items 9 and 4) and only exceeded 67% on 2 items (Items 1 and 6), whereas patient-item scores ranged from 60% (Item 8) to 97% (Item 7). Three paired differences exceeded 50% (Items 9, 4, and 7) and 3 were below 20% (Items 6, 8 and 1), underlying the lack of correlation between peer and patient scores.

 

DISCUSSION

In this multisite study of SDM during inpatient attending rounds, SDM quality, specific SDM behaviors, and factors contributing to SDM were identified. Our study found an adjusted overall Peer-RPAD Score of 4.4 out of 9, and found the following 3 SDM elements most needing improvement according to trained peer observers: (1) “Checking understanding of the patient’s perspective”, (2) “Examining barriers to follow-through with the treatment plan”, and (3) “Asking if the patient has questions.” Areas of strength included explaining the clinical issue or nature of the decision and matching medical language to the patient’s level of understanding, with each rated highly by both peer-observers and patients. Broadly speaking, physicians were skillful in delivering information to patients but failed to solicit input from patients. Characteristics associated with increased SDM in the multivariate analysis included the following: service, patient gender, timing of rounds during patient’s hospital stay, and amount of time rounding with each patient.

Patients similarly found that physicians could improve their abilities to elicit information from patients and families, noting the 3 lowest patient-rated SDM elements were as follows: (1) asking open-ended questions, (2) discussing alternatives or uncertainties, and (3) discussing barriers to treatment plan follow through. Overall, patients and guardians perceived the quantity and quality of SDM on rounds more favorably than peer observers, which is consistent with other studies of patient perceptions of communication. ^29-31 It is possible that patient ratings are more influenced by demand characteristics, fear of negatively impacting their patient-provider relationships, and conflation of overall satisfaction with quality of communication.³² This difference in patient perception of SDM is worthy of further study.

Prior work has revealed that SDM may occur infrequently during inpatient rounds.¹¹ This study further elucidates specific SDM behaviors used along with univariate and multivariate modeling to explore possible contributing factors. The strengths and weaknesses found were similar at all 4 services and the influence of the service was more important than variability across attendings. This study’s findings are similar to a study by Shields et al.,³³ in which the findings in a geographically different outpatient setting 10 years earlier suggesting global and enduring challenges to SDM. To our knowledge, this is the first published study to characterize inpatient SDM behaviors and may serve as the basis for future interventions.

Although the item-level components were ranked similarly across services, on average the summary Peer-RPAD score was lowest at Med-2, where we observed high variability within and between attendings, and was highest at Med-1, where variability was low. Med-2 carried the highest caseload and held the longest rounds, while Med-1 carried the lowest caseload, suggesting that modifiable burdens may hamper SDM performance. Prior studies suggest that patients are often selected based on teaching opportunities, immediate medical need and being newly admitted.³⁴ The high scores at Med-1 may reflect that service’s prediscussion of patients during card-flipping rounds or their selection of which patients to round on as a team. Consistent with prior studies^29,35 of SDM and the family-centered rounding model, which includes the involvement of nurses, respiratory therapists, pharmacists, case managers, social workers, and interpreters on rounds, both pediatrics services showed higher SDM scores.

In contrast to prior studies,^34,36 team size and number of learners did not affect SDM performance, nor did decision type. Despite teams having up to 17 members, 8 learners, and 14 complex patients, SDM scores did not vary significantly by team. Nonetheless, trends were in the directions expected: Scores tended to decrease as the team size or the percentage of trainees grew, and increased with the seniority of the presenting physician. Interestingly, SDM performance decreased with round-average minutes per patient, which may be measuring on-going intensity across cases that leads to exhaustion. Statistically significant patient factors for increased SDM included longer duration of patient encounters, second week of hospital stay, and female patient gender. Although we anticipated that the high number of decisions made early in hospitalization would facilitate higher SDM scores, continuity and stronger patient-provider relationships may enhance SDM.³⁶ We report service-specific team and patient characteristics, in addition to SDM findings in anticipation that some readers will identify with 1 service more than others.

This study has several important limitations. First, our peer observers were not blinded and primarily observed encounters at their own site. To minimize bias, observers periodically rated videos to recalibrate RPAD scoring. Second, additional SDM conversations with a patient and/or guardian may have occurred outside of rounds and were not captured, and poor patient recall may have affected Patient-RPAD scores despite interviewer prompts and timeliness of interviews within 12 hours of rounds. Third, there might have been a selection bias for the one service who selected a smaller number of patients to see, compared with the three other services that performed bedside rounds on all patients. It is possible that attending physicians selected patients who were deemed most able to have SDM conversations, thus affecting RPAD scores on that service. Fourth, study services had fewer patients on average than other academic hospitals (median 9, range 3-14), which might limit its generalizability. Last, as in any observational study, there is always the possibility of the Hawthorne effect. However, neither teams nor patients knew the study objectives.

Nevertheless, important findings emerged through the use of RPAD Scores to evaluate inpatient SDM practices. In particular, we found that to increase SDM quality in inpatient settings, practitioners should (1) check their understanding of the patient’s perspective, (2) examine barriers to follow-through with the treatment plan, and (3) ask if the patient has questions. Variation among services remained very influential after adjusting for team and patient characteristics, which suggests that “climate” or service culture should be targeted by an intervention, rather than individual attendings or subgroups defined by team or patient characteristics. Notably, team size, number of learners, patient census, and type of decision being made did not affect SDM performance, suggesting that even large, busy services can perform SDM if properly trained.

 

Acknowledgments

The authors thank the patients, families, pediatric and internal medicine residents, and hospitalists at Stanford School of Medicine and University of California, San Francisco School of Medicine for their participation in this study. We would also like to thank the student volunteers who collected patient perspectives on the encounters.

Disclosure 

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was supported by an NIH/NCCIH grant R25 AT006573.

Evolving models of medical care emphasize the importance of shared decision-making (SDM) on practical and ethical grounds.^1-3 SDM is a cognitive, emotional, and relational process in which provider and patient collaborate in a decision after discussing the options, evidence, and potential benefits and harms, while considering the patient’s values, preferences, and circumstances.⁴ Categories of decisions include information gathering, pharmacotherapy, therapeutic procedures, consultations and referrals, counseling and precautions (eg, behavior modification, goals of care, end-of-life care), and care transitions (eg, transfer or discharge to home).⁵ Decisions span the continuum of urgency and may be anticipatory or reactive.⁶ The patient’s environment^7,8 and the provider-patient relationship⁹ have been explicitly incorporated into the ideal SDM process.

SDM has been conceptually and empirically linked with evidence-based practice,¹ although the relationship between SDM and clinical outcomes is less clear.^10,11 SDM is desired by patients¹² and may bolster patient satisfaction, trust, and adherence.^13,14 Limited evidence suggests SDM could reduce inappropriate treatments and testing,¹⁵ decrease adverse events,¹⁶ and promote greater patient safety,^17-19 but more well-designed studies are needed.

Provider, patient, and contextual factors influence the extent to which SDM occurs. Providers commonly cite time constraints and perceived lack of applicability to certain clinical scenarios or settings.¹⁹ Providers may also lack training and competency in SDM skills.² Patients may be reluctant to disagree with their provider or fear being mislabeled as “difficult.”²⁰ When faced with high stakes or emotionally charged decisions, patients’ surrogates may prefer to have the provider serve as the sole decision-maker.²¹ Contextually, there may be limited evidence, high clinical stake, or a number of equally beneficial (or harmful) options.^22,23

Current SDM models guide clinicians in determining when and how to engage in SDM, yet models vary widely. For example, Elwyn’s model emphasizes the ethical imperative for SDM and outlines 3 SDM steps: introduce choice, describe options, and help patients explore preferences and make decisions.³ Using a multimodal review and clinician-driven feedback, Legaré’s “IP-SDM” (Interprofessional Shared Decision Making) model illustrates the roles of the interprofessional team and emphasizes the influence of environmental factors on decision-making.²⁴ Recent systematic reviews of SDM models have attempted to identify common elements, language, and processes.^2,25,26

Although published SDM models demonstrate varying emphases–eg, evidence-based medicine,² provider-patient relationships,⁹ interprofessional practices and environmental influences,²⁴ or patient contextual factors ^7,8–none specifically address hospitalization and the issues that impact decisions as a patients’ clinical condition and care needs change. Studies of SDM in hospitalized patients have relied on either general theoretical frameworks for patient engagement or conceptual models developed specifically for outpatient care.^16,27,28 Although the key tenets of SDM are relevant across clinical settings, hospitalization introduces a number of unique and highly relevant factors that may influence all aspects of the SDM process. Table 1 provides several examples from the authors of how inpatient and outpatient SDM may differ.

This study reviews leading SDM models to construct a more environmentally and contextually sensitive model that is appropriate for the hospital setting. Although developed with hospital medicine in mind, a synthesized model that attends to environmental and systems context, provider/team factors, patient factors, and disease/medical variables is highly relevant in any setting where SDM occurs.

METHODS

We constructed a model that is appropriate for SDM across the care continuum through the following 3-part, iterative group process: (1) a comprehensive literature review of existing SDM models, (2) synthesis and inductive development of a new draft model, and (3) modification of the new model using feedback from SDM experts.

Narrative Literature Review

We performed a structured, comprehensive literature review ²⁹ to compare and contrast existing SDM models and frameworks. Leading models and key concepts were first identified using 2 systematic reviews ^25,26 and a comprehensive review.² In order to extend the search to 2016 and include any overlooked articles, a PubMed search was performed using the terms “shared decision-making” or “medical decision-making” AND “model” or “theory” or “framework” for English-language articles from inception to 2016. The search was repeated using Google Scholar to verify results and obtain the number of citations per article as a proxy for impact and saturation. In order to minimize possible search error or selection bias, reference lists in high-impact publications were hand searched to identify additional articles. All abstracts were manually reviewed by 2 independent authors for relevance and later inclusion in our group iterative process. A priori inclusion criteria were limited to provider-patient SDM (ie, not clinical reasoning or making decisions in general) and complete descriptions of a conceptual model or framework. Additional publications suggested by experts (eg, perspective pieces or terminology summaries) were also reviewed.

 

Model Development and Expert Review

An electronic SDM reference library and annotated bibliography of the selected articles (Table 2) was created to guide the synthesis of SDM models and highlight needed revisions for hospital medicine. In a process similar to Legaré,²⁴ a group of 8 pediatric and adult medicine hospitalists, a palliative care physician, a cognitive psychologist, a biostatistician, and 3 medical trainees reviewed the selected SDM publications and models³⁰ and independently created their own adapted inpatient SDM models. Through an iterative, consensus-building group process, each model was discussed to select key elements or features to be integrated into a synthesized model. This model was guided by principles of social ecological theory, which emphasizes the role of the individual as influenced by and interactive with systems and the environment.³¹

The draft model and a standardized set of questions (supplementary Appendix A) were then emailed to all first and last authors of the reviewed studies (Table 2). Expert responses were compiled, coded, and analyzed independently by 3 coauthors. Inductive coding techniques and a constant comparative approach were used to code the qualitative data.³² Preliminary findings were shared among the 3 reviewers and discussed until consensus was reached on emerging themes and implications for the new SDM model and multistep SDM pathway. A master list of suggested revisions was shared with the larger authorship team and the model was refined accordingly.

RESULTS

Two previously published systematic reviews^25,26 identified 494 articles, 161 conceptual definitions of SDM, and over 30 separate key concepts. The additional PubMed search garnered 1957 publications (with many overlapping from the systematic reviews). A manual search of the systematic reviews and PubMed abstracts identified 16 unique and complete decision-making models for further review. Hand searches of their citations yielded an additional 6 models for a total of 22 models.^{3,4,13,23,33-51} The majority of excluded articles described specific decision aids and small clinical studies, focused on only one step of the decision-making process, or were not otherwise relevant. The first (SR) and senior authors (JS) reviewed the 22 models for SDM relevance, generalizability, and content saturation, yielding a final sample of 9 SDM models. A subsequent Google Scholar search did not identify any new SDM models but 2 SDM theory papers^1,52 and 2 commentaries^53,54 were selected based on influence (ie, number of citations), expert recommendation, or coverage of a novel aspect of SDM. A total of 15 studies (9 SDM models + 6 reviews; Table 2) were used by our development team to create a synthesized SDM model. A 10th SDM model⁵⁵ and 3 additional descriptive and normative studies^8,56,57 were later added based on expert feedback and incorporated into our final SDM 3 Circle Model.

Expert Feedback

Twenty-one of 27 (78%) SDM expert authors responded to our e-mail request for feedback. The majority (62%) agreed with the basic elements of the model, including the environmental frame and the 3 domains. Some respondents viewed SDM as strictly a process between patient and provider independent of the disease, leading to refinement of the medical context category. Several experts emphasized the importance of SDM “set-up,” which includes the elicitation of patient preferences in how decisions are made and the extent of patient and/or surrogate involvement.

Several respondents identified time constraints (N = 2), acuity of disease (N = 3), and presence of multiple teams (N = 6) to be the significant factors distinguishing inpatient from outpatient SDM. For some experts, “team” referred to the interprofessional care team, whereas others referred to it as the collaboration among attending physicians and trainees. Experts noted that although the intensity and frequency of inpatient interactions could promote SDM, higher patient acuity and the urgency of decisions could negatively influence SDM and/or the patient’s ability to participate. Similarly, the presence of other team members may either impede or promote SDM by either contributing to miscommunication or bringing well-trained SDM experts to the bedside. Financial impact on patients and resource constraints were also noted as relevant. All of these elements have been incorporated into the final SDM 3 Circle Model and multistep SDM Pathway (Supplemental Appendix A and B).

The SDM 3 Circle Model

The SDM 3 Circle Model comprises 3 categories of SDM barriers and facilitators that intersect within the environmental frame of an inpatient ward or other setting: (1) provider/team, (2) patient/family, and (3) medical context. A Venn diagram visually represents the conceptual overlaps and distinctions among these categories that are all affected by the environment in which they occur (Supplemental Appendix A).

The patient/family circle mirrors prior SDM models that address the role of patient preferences in making decisions,^3,4,12 with the explicit addition of the roles of families and surrogates as either decision-makers or influencers. This circle includes personal characteristics, such as cognitions (eg, beliefs, attitudes), emotions (eg, anxiety, hope), behaviors (eg, adherence, assertiveness), illness history (ie, subjective experience and understanding of one’s own medical history), and related social features (eg, culture, education, literacy, social supports).

Patient factors are not static over time or context. They occur within an environmental setting and are likely to be influenced by concurrent provider and medical variables (the second and third circles). Disease exacerbation leading to hospitalization or transfer to a subacute facility could dramatically shift the calculus a patient uses to determine preferences or activate dormant family dynamics. Strong provider-patient rapport (the overlap of patient and provider factors) may influence the development of trust and subsequent decisions.⁹ The type of disease or symptom presentation (circle 3–medical context) may further influence patient factors due to stigma, perceived vulnerability, or assumed prognosis.

The provider/team circle includes both individual and team-based factors falling into similar categories as the patient/family domain, such as cognitions, behavior, and social features; however, these factors include both personal (eg, the provider’s personal history, values, and beliefs) and professional (eg, past medical training, decision-making style, past experiences treating a disease) characteristics. Decisions may involve an interprofessional team representing a broad range of personalities and professional values. Decisions and decision-making processes may change over time as team composition changes, as level of provider expertise varies, or as environmental, patient, or disease/illness factors influence providers and teams.

Medical context includes factors related to the disease and the potential ways to evaluate or manage it. Examples of disease factors include acuity, symptoms, course, and prognosis. Most obviously, disease factors will influence the content of risk-benefit discussions but may also affect the SDM process through disease stigma or cultural assumptions about etiology. Disease evaluation factors include the psychometrics of a diagnostic screen, invasive and noninvasive testing, or a range of different preventive or therapeutic interventions. Treatment variables include the available options, costs, and risk of complications. Medical context variables evolve as evidence-based medicine and biomedical knowledge increase and new treatment options emerge.

Each of the 3 circles operates within the same environmental frame, such as an inpatient medicine ward, which itself operates within a hospital and the broader healthcare system. This frame exerts overt and subtle influences on providers, patients, and even the medical context. Features of the environmental frame include culture (eg, values, preferences, social norms), university versus community setting, incentives, formularies, quality improvement campaigns, regulations, and technology use.

The dynamic interactivity of the environmental frame and the 3 circles inform the process of SDM and highlight key differences that may occur between care settings. Certain features may predominate in different situations, but all will influence and be influenced by features of other circles during the course of SDM.

 

Application of the SDM 3 Circle Model

As shown in the Figure, the multistep SDM pathway begins with information gathering and processing, where the provider solicits medical history as well as patient preferences for decision-making. This “processing” of patient decision-making preferences is less commonly described. The next steps, sharing information and decision discussion, include patient education about the medical issue and available treatments. Discussions may involve the pros/cons of each option, alternative diagnostic or management strategies, and how these decisions fit with a patient’s preferences, abilities (eg, health literacy)⁵⁸ and resources, or what has been called “contextualizing care.”^7,8 Framing and other provider behaviors, including the use of decision aids and decision guides,¹⁵ may influence these conversations. Finally, after gathering, sharing, and discussing information (as influenced by the environment and 3 circles), a medical decision is made and patient understanding is verified. Detailed examples of how this model might be applied are illustrated with case scenarios in supplemental Appendix B.

Although the SDM process is similar across clinical settings, its operationalization varies in important ways for hospital decision-making. In some situations, patients may defer all decisions to their providers or decisions may be considered with multiple providers concurrently. In the hospital, SDM may not be possible, such as in emergency surgery for an obtunded patient or when the patient and surrogate are not available or able to participate in the decision. Therefore, providers may bypass the steps of information sharing and discussion of the decision (big arrow in the Figure and supplemental Appendix B), proceeding directly to decision making.

DISCUSSION

The SDM 3 Circle Model provides a concise, ecologically valid, contextually sensitive representation of SDM that synthesizes and extends beyond recent SDM models.^3,7,40 Each circle represents the forces that influence SDM across settings. Although the multistep SDM pathway occurs similarly in outpatient and inpatient settings, how each step is operationalized and how each “circle” exerts its influence may differ and warrants further consideration throughout the SDM process. For example, hospitalized patients may have greater stress and anxiety, have more family involvement, be more motivated to adhere to treatment, and may be under greater financial and social pressures. Unlike outpatient primary care, patients are less likely to have an existing relationship with their inpatient providers, potentially compromising patient confidence in the provider, and necessitating expeditious trust building.

The SDM 3 Circle Model captures “setting” in both the broader environmental frame and within the provider/team category of variables. The frame also captures health system and broader community variables that may influence the practicality of some medical decisions. Within this essential frame, all 3 categories of patient, provider, and medical context are included as part of the SDM process. A better understanding of their interplay may be of great value for clinicians, researchers, administrators, and policy makers who wish to further study and promote SDM. Both the SDM 3 Circle Model and its accompanying pathway (Figures 1 and 2) highlight opportunities for intervention and research, and may drive quality improvement initiatives to improve clinical outcomes.

Limitations

We did not perform a new systematic review, potentially omitting lesser-known publications. We mitigated this risk by using recent systematic reviews, searching multiple databases, hand searching citation lists, and making inquiries to SDM experts. Our selection of models used as a foundation for the synthesized model was based on consensus, which included an element of subjective, clinical judgment. Our SDM expert sample was small and limited to authors of the papers we reviewed, potentially restricting the range of viewpoints received. Lastly, the SDM 3 Circle Model highlights key concept areas rather than all possible factors that influence SDM.

CONCLUSIONS

We present a peer-reviewed, literature-based SDM model capable of accounting for the unique circumstances and challenges of SDM in the hospital. The SDM 3 Circle Model identifies the primary categories of variables thought to influence SDM, places them in a shared environmental frame, and visually represents their interactive nature. A multistep representation of the SDM process further illustrates how the unique features and challenges of hospitalization might exert influence at various points as patients and providers reach a shared decision. As the interrelationships of patient and provider/team, medical context, and the environmental frame in which they occur are better understood, more effective and targeted interventions to promote SDM can be developed and evaluated.

Acknowledgments

The authors would like to thank Evans Whitaker for his assistance with the literature review and the Patient Engagement Project volunteers for their support and assistance with data collection.

Disclosure

Financial support for this study was provided entirely by a grant from NIH/NCCIH (grant #R25 AT006573, awarded to Dr. Jason Satterfield). The funding agreement ensured the authors’ independence in designing the study, interpreting the data, writing, and publishing the report. The following authors are employed by the sponsor: Stephanie Rennke, MD, Patrick Yuan, BA, Brad Monash, MD, Rebecca Blankenburg, MD, MPH, Ian Chua, MD, Stephanie Harman, MD, Debbie S. Sakai, MD, Joan F. Hilton, DSc, MPH., and Jason Satterfield, PhD.

 

Evolving models of medical care emphasize the importance of shared decision-making (SDM) on practical and ethical grounds.^1-3 SDM is a cognitive, emotional, and relational process in which provider and patient collaborate in a decision after discussing the options, evidence, and potential benefits and harms, while considering the patient’s values, preferences, and circumstances.⁴ Categories of decisions include information gathering, pharmacotherapy, therapeutic procedures, consultations and referrals, counseling and precautions (eg, behavior modification, goals of care, end-of-life care), and care transitions (eg, transfer or discharge to home).⁵ Decisions span the continuum of urgency and may be anticipatory or reactive.⁶ The patient’s environment^7,8 and the provider-patient relationship⁹ have been explicitly incorporated into the ideal SDM process.

SDM has been conceptually and empirically linked with evidence-based practice,¹ although the relationship between SDM and clinical outcomes is less clear.^10,11 SDM is desired by patients¹² and may bolster patient satisfaction, trust, and adherence.^13,14 Limited evidence suggests SDM could reduce inappropriate treatments and testing,¹⁵ decrease adverse events,¹⁶ and promote greater patient safety,^17-19 but more well-designed studies are needed.

Provider, patient, and contextual factors influence the extent to which SDM occurs. Providers commonly cite time constraints and perceived lack of applicability to certain clinical scenarios or settings.¹⁹ Providers may also lack training and competency in SDM skills.² Patients may be reluctant to disagree with their provider or fear being mislabeled as “difficult.”²⁰ When faced with high stakes or emotionally charged decisions, patients’ surrogates may prefer to have the provider serve as the sole decision-maker.²¹ Contextually, there may be limited evidence, high clinical stake, or a number of equally beneficial (or harmful) options.^22,23

Current SDM models guide clinicians in determining when and how to engage in SDM, yet models vary widely. For example, Elwyn’s model emphasizes the ethical imperative for SDM and outlines 3 SDM steps: introduce choice, describe options, and help patients explore preferences and make decisions.³ Using a multimodal review and clinician-driven feedback, Legaré’s “IP-SDM” (Interprofessional Shared Decision Making) model illustrates the roles of the interprofessional team and emphasizes the influence of environmental factors on decision-making.²⁴ Recent systematic reviews of SDM models have attempted to identify common elements, language, and processes.^2,25,26

Although published SDM models demonstrate varying emphases–eg, evidence-based medicine,² provider-patient relationships,⁹ interprofessional practices and environmental influences,²⁴ or patient contextual factors ^7,8–none specifically address hospitalization and the issues that impact decisions as a patients’ clinical condition and care needs change. Studies of SDM in hospitalized patients have relied on either general theoretical frameworks for patient engagement or conceptual models developed specifically for outpatient care.^16,27,28 Although the key tenets of SDM are relevant across clinical settings, hospitalization introduces a number of unique and highly relevant factors that may influence all aspects of the SDM process. Table 1 provides several examples from the authors of how inpatient and outpatient SDM may differ.

This study reviews leading SDM models to construct a more environmentally and contextually sensitive model that is appropriate for the hospital setting. Although developed with hospital medicine in mind, a synthesized model that attends to environmental and systems context, provider/team factors, patient factors, and disease/medical variables is highly relevant in any setting where SDM occurs.

METHODS

We constructed a model that is appropriate for SDM across the care continuum through the following 3-part, iterative group process: (1) a comprehensive literature review of existing SDM models, (2) synthesis and inductive development of a new draft model, and (3) modification of the new model using feedback from SDM experts.

Narrative Literature Review

We performed a structured, comprehensive literature review ²⁹ to compare and contrast existing SDM models and frameworks. Leading models and key concepts were first identified using 2 systematic reviews ^25,26 and a comprehensive review.² In order to extend the search to 2016 and include any overlooked articles, a PubMed search was performed using the terms “shared decision-making” or “medical decision-making” AND “model” or “theory” or “framework” for English-language articles from inception to 2016. The search was repeated using Google Scholar to verify results and obtain the number of citations per article as a proxy for impact and saturation. In order to minimize possible search error or selection bias, reference lists in high-impact publications were hand searched to identify additional articles. All abstracts were manually reviewed by 2 independent authors for relevance and later inclusion in our group iterative process. A priori inclusion criteria were limited to provider-patient SDM (ie, not clinical reasoning or making decisions in general) and complete descriptions of a conceptual model or framework. Additional publications suggested by experts (eg, perspective pieces or terminology summaries) were also reviewed.

 

Model Development and Expert Review

An electronic SDM reference library and annotated bibliography of the selected articles (Table 2) was created to guide the synthesis of SDM models and highlight needed revisions for hospital medicine. In a process similar to Legaré,²⁴ a group of 8 pediatric and adult medicine hospitalists, a palliative care physician, a cognitive psychologist, a biostatistician, and 3 medical trainees reviewed the selected SDM publications and models³⁰ and independently created their own adapted inpatient SDM models. Through an iterative, consensus-building group process, each model was discussed to select key elements or features to be integrated into a synthesized model. This model was guided by principles of social ecological theory, which emphasizes the role of the individual as influenced by and interactive with systems and the environment.³¹

The draft model and a standardized set of questions (supplementary Appendix A) were then emailed to all first and last authors of the reviewed studies (Table 2). Expert responses were compiled, coded, and analyzed independently by 3 coauthors. Inductive coding techniques and a constant comparative approach were used to code the qualitative data.³² Preliminary findings were shared among the 3 reviewers and discussed until consensus was reached on emerging themes and implications for the new SDM model and multistep SDM pathway. A master list of suggested revisions was shared with the larger authorship team and the model was refined accordingly.

RESULTS

Two previously published systematic reviews^25,26 identified 494 articles, 161 conceptual definitions of SDM, and over 30 separate key concepts. The additional PubMed search garnered 1957 publications (with many overlapping from the systematic reviews). A manual search of the systematic reviews and PubMed abstracts identified 16 unique and complete decision-making models for further review. Hand searches of their citations yielded an additional 6 models for a total of 22 models.^{3,4,13,23,33-51} The majority of excluded articles described specific decision aids and small clinical studies, focused on only one step of the decision-making process, or were not otherwise relevant. The first (SR) and senior authors (JS) reviewed the 22 models for SDM relevance, generalizability, and content saturation, yielding a final sample of 9 SDM models. A subsequent Google Scholar search did not identify any new SDM models but 2 SDM theory papers^1,52 and 2 commentaries^53,54 were selected based on influence (ie, number of citations), expert recommendation, or coverage of a novel aspect of SDM. A total of 15 studies (9 SDM models + 6 reviews; Table 2) were used by our development team to create a synthesized SDM model. A 10th SDM model⁵⁵ and 3 additional descriptive and normative studies^8,56,57 were later added based on expert feedback and incorporated into our final SDM 3 Circle Model.

Expert Feedback

Twenty-one of 27 (78%) SDM expert authors responded to our e-mail request for feedback. The majority (62%) agreed with the basic elements of the model, including the environmental frame and the 3 domains. Some respondents viewed SDM as strictly a process between patient and provider independent of the disease, leading to refinement of the medical context category. Several experts emphasized the importance of SDM “set-up,” which includes the elicitation of patient preferences in how decisions are made and the extent of patient and/or surrogate involvement.

Several respondents identified time constraints (N = 2), acuity of disease (N = 3), and presence of multiple teams (N = 6) to be the significant factors distinguishing inpatient from outpatient SDM. For some experts, “team” referred to the interprofessional care team, whereas others referred to it as the collaboration among attending physicians and trainees. Experts noted that although the intensity and frequency of inpatient interactions could promote SDM, higher patient acuity and the urgency of decisions could negatively influence SDM and/or the patient’s ability to participate. Similarly, the presence of other team members may either impede or promote SDM by either contributing to miscommunication or bringing well-trained SDM experts to the bedside. Financial impact on patients and resource constraints were also noted as relevant. All of these elements have been incorporated into the final SDM 3 Circle Model and multistep SDM Pathway (Supplemental Appendix A and B).

The SDM 3 Circle Model

The SDM 3 Circle Model comprises 3 categories of SDM barriers and facilitators that intersect within the environmental frame of an inpatient ward or other setting: (1) provider/team, (2) patient/family, and (3) medical context. A Venn diagram visually represents the conceptual overlaps and distinctions among these categories that are all affected by the environment in which they occur (Supplemental Appendix A).

The patient/family circle mirrors prior SDM models that address the role of patient preferences in making decisions,^3,4,12 with the explicit addition of the roles of families and surrogates as either decision-makers or influencers. This circle includes personal characteristics, such as cognitions (eg, beliefs, attitudes), emotions (eg, anxiety, hope), behaviors (eg, adherence, assertiveness), illness history (ie, subjective experience and understanding of one’s own medical history), and related social features (eg, culture, education, literacy, social supports).

Patient factors are not static over time or context. They occur within an environmental setting and are likely to be influenced by concurrent provider and medical variables (the second and third circles). Disease exacerbation leading to hospitalization or transfer to a subacute facility could dramatically shift the calculus a patient uses to determine preferences or activate dormant family dynamics. Strong provider-patient rapport (the overlap of patient and provider factors) may influence the development of trust and subsequent decisions.⁹ The type of disease or symptom presentation (circle 3–medical context) may further influence patient factors due to stigma, perceived vulnerability, or assumed prognosis.

The provider/team circle includes both individual and team-based factors falling into similar categories as the patient/family domain, such as cognitions, behavior, and social features; however, these factors include both personal (eg, the provider’s personal history, values, and beliefs) and professional (eg, past medical training, decision-making style, past experiences treating a disease) characteristics. Decisions may involve an interprofessional team representing a broad range of personalities and professional values. Decisions and decision-making processes may change over time as team composition changes, as level of provider expertise varies, or as environmental, patient, or disease/illness factors influence providers and teams.

Medical context includes factors related to the disease and the potential ways to evaluate or manage it. Examples of disease factors include acuity, symptoms, course, and prognosis. Most obviously, disease factors will influence the content of risk-benefit discussions but may also affect the SDM process through disease stigma or cultural assumptions about etiology. Disease evaluation factors include the psychometrics of a diagnostic screen, invasive and noninvasive testing, or a range of different preventive or therapeutic interventions. Treatment variables include the available options, costs, and risk of complications. Medical context variables evolve as evidence-based medicine and biomedical knowledge increase and new treatment options emerge.

Each of the 3 circles operates within the same environmental frame, such as an inpatient medicine ward, which itself operates within a hospital and the broader healthcare system. This frame exerts overt and subtle influences on providers, patients, and even the medical context. Features of the environmental frame include culture (eg, values, preferences, social norms), university versus community setting, incentives, formularies, quality improvement campaigns, regulations, and technology use.

The dynamic interactivity of the environmental frame and the 3 circles inform the process of SDM and highlight key differences that may occur between care settings. Certain features may predominate in different situations, but all will influence and be influenced by features of other circles during the course of SDM.

 

Application of the SDM 3 Circle Model

As shown in the Figure, the multistep SDM pathway begins with information gathering and processing, where the provider solicits medical history as well as patient preferences for decision-making. This “processing” of patient decision-making preferences is less commonly described. The next steps, sharing information and decision discussion, include patient education about the medical issue and available treatments. Discussions may involve the pros/cons of each option, alternative diagnostic or management strategies, and how these decisions fit with a patient’s preferences, abilities (eg, health literacy)⁵⁸ and resources, or what has been called “contextualizing care.”^7,8 Framing and other provider behaviors, including the use of decision aids and decision guides,¹⁵ may influence these conversations. Finally, after gathering, sharing, and discussing information (as influenced by the environment and 3 circles), a medical decision is made and patient understanding is verified. Detailed examples of how this model might be applied are illustrated with case scenarios in supplemental Appendix B.

Although the SDM process is similar across clinical settings, its operationalization varies in important ways for hospital decision-making. In some situations, patients may defer all decisions to their providers or decisions may be considered with multiple providers concurrently. In the hospital, SDM may not be possible, such as in emergency surgery for an obtunded patient or when the patient and surrogate are not available or able to participate in the decision. Therefore, providers may bypass the steps of information sharing and discussion of the decision (big arrow in the Figure and supplemental Appendix B), proceeding directly to decision making.

DISCUSSION

The SDM 3 Circle Model provides a concise, ecologically valid, contextually sensitive representation of SDM that synthesizes and extends beyond recent SDM models.^3,7,40 Each circle represents the forces that influence SDM across settings. Although the multistep SDM pathway occurs similarly in outpatient and inpatient settings, how each step is operationalized and how each “circle” exerts its influence may differ and warrants further consideration throughout the SDM process. For example, hospitalized patients may have greater stress and anxiety, have more family involvement, be more motivated to adhere to treatment, and may be under greater financial and social pressures. Unlike outpatient primary care, patients are less likely to have an existing relationship with their inpatient providers, potentially compromising patient confidence in the provider, and necessitating expeditious trust building.

The SDM 3 Circle Model captures “setting” in both the broader environmental frame and within the provider/team category of variables. The frame also captures health system and broader community variables that may influence the practicality of some medical decisions. Within this essential frame, all 3 categories of patient, provider, and medical context are included as part of the SDM process. A better understanding of their interplay may be of great value for clinicians, researchers, administrators, and policy makers who wish to further study and promote SDM. Both the SDM 3 Circle Model and its accompanying pathway (Figures 1 and 2) highlight opportunities for intervention and research, and may drive quality improvement initiatives to improve clinical outcomes.

Limitations

We did not perform a new systematic review, potentially omitting lesser-known publications. We mitigated this risk by using recent systematic reviews, searching multiple databases, hand searching citation lists, and making inquiries to SDM experts. Our selection of models used as a foundation for the synthesized model was based on consensus, which included an element of subjective, clinical judgment. Our SDM expert sample was small and limited to authors of the papers we reviewed, potentially restricting the range of viewpoints received. Lastly, the SDM 3 Circle Model highlights key concept areas rather than all possible factors that influence SDM.

CONCLUSIONS

We present a peer-reviewed, literature-based SDM model capable of accounting for the unique circumstances and challenges of SDM in the hospital. The SDM 3 Circle Model identifies the primary categories of variables thought to influence SDM, places them in a shared environmental frame, and visually represents their interactive nature. A multistep representation of the SDM process further illustrates how the unique features and challenges of hospitalization might exert influence at various points as patients and providers reach a shared decision. As the interrelationships of patient and provider/team, medical context, and the environmental frame in which they occur are better understood, more effective and targeted interventions to promote SDM can be developed and evaluated.

Acknowledgments

The authors would like to thank Evans Whitaker for his assistance with the literature review and the Patient Engagement Project volunteers for their support and assistance with data collection.

Disclosure

Financial support for this study was provided entirely by a grant from NIH/NCCIH (grant #R25 AT006573, awarded to Dr. Jason Satterfield). The funding agreement ensured the authors’ independence in designing the study, interpreting the data, writing, and publishing the report. The following authors are employed by the sponsor: Stephanie Rennke, MD, Patrick Yuan, BA, Brad Monash, MD, Rebecca Blankenburg, MD, MPH, Ian Chua, MD, Stephanie Harman, MD, Debbie S. Sakai, MD, Joan F. Hilton, DSc, MPH., and Jason Satterfield, PhD.

 

Evolving models of medical care emphasize the importance of shared decision-making (SDM) on practical and ethical grounds.^1-3 SDM is a cognitive, emotional, and relational process in which provider and patient collaborate in a decision after discussing the options, evidence, and potential benefits and harms, while considering the patient’s values, preferences, and circumstances.⁴ Categories of decisions include information gathering, pharmacotherapy, therapeutic procedures, consultations and referrals, counseling and precautions (eg, behavior modification, goals of care, end-of-life care), and care transitions (eg, transfer or discharge to home).⁵ Decisions span the continuum of urgency and may be anticipatory or reactive.⁶ The patient’s environment^7,8 and the provider-patient relationship⁹ have been explicitly incorporated into the ideal SDM process.

SDM has been conceptually and empirically linked with evidence-based practice,¹ although the relationship between SDM and clinical outcomes is less clear.^10,11 SDM is desired by patients¹² and may bolster patient satisfaction, trust, and adherence.^13,14 Limited evidence suggests SDM could reduce inappropriate treatments and testing,¹⁵ decrease adverse events,¹⁶ and promote greater patient safety,^17-19 but more well-designed studies are needed.

Provider, patient, and contextual factors influence the extent to which SDM occurs. Providers commonly cite time constraints and perceived lack of applicability to certain clinical scenarios or settings.¹⁹ Providers may also lack training and competency in SDM skills.² Patients may be reluctant to disagree with their provider or fear being mislabeled as “difficult.”²⁰ When faced with high stakes or emotionally charged decisions, patients’ surrogates may prefer to have the provider serve as the sole decision-maker.²¹ Contextually, there may be limited evidence, high clinical stake, or a number of equally beneficial (or harmful) options.^22,23

Current SDM models guide clinicians in determining when and how to engage in SDM, yet models vary widely. For example, Elwyn’s model emphasizes the ethical imperative for SDM and outlines 3 SDM steps: introduce choice, describe options, and help patients explore preferences and make decisions.³ Using a multimodal review and clinician-driven feedback, Legaré’s “IP-SDM” (Interprofessional Shared Decision Making) model illustrates the roles of the interprofessional team and emphasizes the influence of environmental factors on decision-making.²⁴ Recent systematic reviews of SDM models have attempted to identify common elements, language, and processes.^2,25,26

Although published SDM models demonstrate varying emphases–eg, evidence-based medicine,² provider-patient relationships,⁹ interprofessional practices and environmental influences,²⁴ or patient contextual factors ^7,8–none specifically address hospitalization and the issues that impact decisions as a patients’ clinical condition and care needs change. Studies of SDM in hospitalized patients have relied on either general theoretical frameworks for patient engagement or conceptual models developed specifically for outpatient care.^16,27,28 Although the key tenets of SDM are relevant across clinical settings, hospitalization introduces a number of unique and highly relevant factors that may influence all aspects of the SDM process. Table 1 provides several examples from the authors of how inpatient and outpatient SDM may differ.

This study reviews leading SDM models to construct a more environmentally and contextually sensitive model that is appropriate for the hospital setting. Although developed with hospital medicine in mind, a synthesized model that attends to environmental and systems context, provider/team factors, patient factors, and disease/medical variables is highly relevant in any setting where SDM occurs.

METHODS

We constructed a model that is appropriate for SDM across the care continuum through the following 3-part, iterative group process: (1) a comprehensive literature review of existing SDM models, (2) synthesis and inductive development of a new draft model, and (3) modification of the new model using feedback from SDM experts.

Narrative Literature Review

We performed a structured, comprehensive literature review ²⁹ to compare and contrast existing SDM models and frameworks. Leading models and key concepts were first identified using 2 systematic reviews ^25,26 and a comprehensive review.² In order to extend the search to 2016 and include any overlooked articles, a PubMed search was performed using the terms “shared decision-making” or “medical decision-making” AND “model” or “theory” or “framework” for English-language articles from inception to 2016. The search was repeated using Google Scholar to verify results and obtain the number of citations per article as a proxy for impact and saturation. In order to minimize possible search error or selection bias, reference lists in high-impact publications were hand searched to identify additional articles. All abstracts were manually reviewed by 2 independent authors for relevance and later inclusion in our group iterative process. A priori inclusion criteria were limited to provider-patient SDM (ie, not clinical reasoning or making decisions in general) and complete descriptions of a conceptual model or framework. Additional publications suggested by experts (eg, perspective pieces or terminology summaries) were also reviewed.

 

Model Development and Expert Review

An electronic SDM reference library and annotated bibliography of the selected articles (Table 2) was created to guide the synthesis of SDM models and highlight needed revisions for hospital medicine. In a process similar to Legaré,²⁴ a group of 8 pediatric and adult medicine hospitalists, a palliative care physician, a cognitive psychologist, a biostatistician, and 3 medical trainees reviewed the selected SDM publications and models³⁰ and independently created their own adapted inpatient SDM models. Through an iterative, consensus-building group process, each model was discussed to select key elements or features to be integrated into a synthesized model. This model was guided by principles of social ecological theory, which emphasizes the role of the individual as influenced by and interactive with systems and the environment.³¹

The draft model and a standardized set of questions (supplementary Appendix A) were then emailed to all first and last authors of the reviewed studies (Table 2). Expert responses were compiled, coded, and analyzed independently by 3 coauthors. Inductive coding techniques and a constant comparative approach were used to code the qualitative data.³² Preliminary findings were shared among the 3 reviewers and discussed until consensus was reached on emerging themes and implications for the new SDM model and multistep SDM pathway. A master list of suggested revisions was shared with the larger authorship team and the model was refined accordingly.

RESULTS

Two previously published systematic reviews^25,26 identified 494 articles, 161 conceptual definitions of SDM, and over 30 separate key concepts. The additional PubMed search garnered 1957 publications (with many overlapping from the systematic reviews). A manual search of the systematic reviews and PubMed abstracts identified 16 unique and complete decision-making models for further review. Hand searches of their citations yielded an additional 6 models for a total of 22 models.^{3,4,13,23,33-51} The majority of excluded articles described specific decision aids and small clinical studies, focused on only one step of the decision-making process, or were not otherwise relevant. The first (SR) and senior authors (JS) reviewed the 22 models for SDM relevance, generalizability, and content saturation, yielding a final sample of 9 SDM models. A subsequent Google Scholar search did not identify any new SDM models but 2 SDM theory papers^1,52 and 2 commentaries^53,54 were selected based on influence (ie, number of citations), expert recommendation, or coverage of a novel aspect of SDM. A total of 15 studies (9 SDM models + 6 reviews; Table 2) were used by our development team to create a synthesized SDM model. A 10th SDM model⁵⁵ and 3 additional descriptive and normative studies^8,56,57 were later added based on expert feedback and incorporated into our final SDM 3 Circle Model.

Expert Feedback

Twenty-one of 27 (78%) SDM expert authors responded to our e-mail request for feedback. The majority (62%) agreed with the basic elements of the model, including the environmental frame and the 3 domains. Some respondents viewed SDM as strictly a process between patient and provider independent of the disease, leading to refinement of the medical context category. Several experts emphasized the importance of SDM “set-up,” which includes the elicitation of patient preferences in how decisions are made and the extent of patient and/or surrogate involvement.

Several respondents identified time constraints (N = 2), acuity of disease (N = 3), and presence of multiple teams (N = 6) to be the significant factors distinguishing inpatient from outpatient SDM. For some experts, “team” referred to the interprofessional care team, whereas others referred to it as the collaboration among attending physicians and trainees. Experts noted that although the intensity and frequency of inpatient interactions could promote SDM, higher patient acuity and the urgency of decisions could negatively influence SDM and/or the patient’s ability to participate. Similarly, the presence of other team members may either impede or promote SDM by either contributing to miscommunication or bringing well-trained SDM experts to the bedside. Financial impact on patients and resource constraints were also noted as relevant. All of these elements have been incorporated into the final SDM 3 Circle Model and multistep SDM Pathway (Supplemental Appendix A and B).

The SDM 3 Circle Model

The SDM 3 Circle Model comprises 3 categories of SDM barriers and facilitators that intersect within the environmental frame of an inpatient ward or other setting: (1) provider/team, (2) patient/family, and (3) medical context. A Venn diagram visually represents the conceptual overlaps and distinctions among these categories that are all affected by the environment in which they occur (Supplemental Appendix A).

The patient/family circle mirrors prior SDM models that address the role of patient preferences in making decisions,^3,4,12 with the explicit addition of the roles of families and surrogates as either decision-makers or influencers. This circle includes personal characteristics, such as cognitions (eg, beliefs, attitudes), emotions (eg, anxiety, hope), behaviors (eg, adherence, assertiveness), illness history (ie, subjective experience and understanding of one’s own medical history), and related social features (eg, culture, education, literacy, social supports).

Patient factors are not static over time or context. They occur within an environmental setting and are likely to be influenced by concurrent provider and medical variables (the second and third circles). Disease exacerbation leading to hospitalization or transfer to a subacute facility could dramatically shift the calculus a patient uses to determine preferences or activate dormant family dynamics. Strong provider-patient rapport (the overlap of patient and provider factors) may influence the development of trust and subsequent decisions.⁹ The type of disease or symptom presentation (circle 3–medical context) may further influence patient factors due to stigma, perceived vulnerability, or assumed prognosis.

The provider/team circle includes both individual and team-based factors falling into similar categories as the patient/family domain, such as cognitions, behavior, and social features; however, these factors include both personal (eg, the provider’s personal history, values, and beliefs) and professional (eg, past medical training, decision-making style, past experiences treating a disease) characteristics. Decisions may involve an interprofessional team representing a broad range of personalities and professional values. Decisions and decision-making processes may change over time as team composition changes, as level of provider expertise varies, or as environmental, patient, or disease/illness factors influence providers and teams.

Medical context includes factors related to the disease and the potential ways to evaluate or manage it. Examples of disease factors include acuity, symptoms, course, and prognosis. Most obviously, disease factors will influence the content of risk-benefit discussions but may also affect the SDM process through disease stigma or cultural assumptions about etiology. Disease evaluation factors include the psychometrics of a diagnostic screen, invasive and noninvasive testing, or a range of different preventive or therapeutic interventions. Treatment variables include the available options, costs, and risk of complications. Medical context variables evolve as evidence-based medicine and biomedical knowledge increase and new treatment options emerge.

Each of the 3 circles operates within the same environmental frame, such as an inpatient medicine ward, which itself operates within a hospital and the broader healthcare system. This frame exerts overt and subtle influences on providers, patients, and even the medical context. Features of the environmental frame include culture (eg, values, preferences, social norms), university versus community setting, incentives, formularies, quality improvement campaigns, regulations, and technology use.

The dynamic interactivity of the environmental frame and the 3 circles inform the process of SDM and highlight key differences that may occur between care settings. Certain features may predominate in different situations, but all will influence and be influenced by features of other circles during the course of SDM.

 

Application of the SDM 3 Circle Model

As shown in the Figure, the multistep SDM pathway begins with information gathering and processing, where the provider solicits medical history as well as patient preferences for decision-making. This “processing” of patient decision-making preferences is less commonly described. The next steps, sharing information and decision discussion, include patient education about the medical issue and available treatments. Discussions may involve the pros/cons of each option, alternative diagnostic or management strategies, and how these decisions fit with a patient’s preferences, abilities (eg, health literacy)⁵⁸ and resources, or what has been called “contextualizing care.”^7,8 Framing and other provider behaviors, including the use of decision aids and decision guides,¹⁵ may influence these conversations. Finally, after gathering, sharing, and discussing information (as influenced by the environment and 3 circles), a medical decision is made and patient understanding is verified. Detailed examples of how this model might be applied are illustrated with case scenarios in supplemental Appendix B.

Although the SDM process is similar across clinical settings, its operationalization varies in important ways for hospital decision-making. In some situations, patients may defer all decisions to their providers or decisions may be considered with multiple providers concurrently. In the hospital, SDM may not be possible, such as in emergency surgery for an obtunded patient or when the patient and surrogate are not available or able to participate in the decision. Therefore, providers may bypass the steps of information sharing and discussion of the decision (big arrow in the Figure and supplemental Appendix B), proceeding directly to decision making.

DISCUSSION

The SDM 3 Circle Model provides a concise, ecologically valid, contextually sensitive representation of SDM that synthesizes and extends beyond recent SDM models.^3,7,40 Each circle represents the forces that influence SDM across settings. Although the multistep SDM pathway occurs similarly in outpatient and inpatient settings, how each step is operationalized and how each “circle” exerts its influence may differ and warrants further consideration throughout the SDM process. For example, hospitalized patients may have greater stress and anxiety, have more family involvement, be more motivated to adhere to treatment, and may be under greater financial and social pressures. Unlike outpatient primary care, patients are less likely to have an existing relationship with their inpatient providers, potentially compromising patient confidence in the provider, and necessitating expeditious trust building.

The SDM 3 Circle Model captures “setting” in both the broader environmental frame and within the provider/team category of variables. The frame also captures health system and broader community variables that may influence the practicality of some medical decisions. Within this essential frame, all 3 categories of patient, provider, and medical context are included as part of the SDM process. A better understanding of their interplay may be of great value for clinicians, researchers, administrators, and policy makers who wish to further study and promote SDM. Both the SDM 3 Circle Model and its accompanying pathway (Figures 1 and 2) highlight opportunities for intervention and research, and may drive quality improvement initiatives to improve clinical outcomes.

Limitations

We did not perform a new systematic review, potentially omitting lesser-known publications. We mitigated this risk by using recent systematic reviews, searching multiple databases, hand searching citation lists, and making inquiries to SDM experts. Our selection of models used as a foundation for the synthesized model was based on consensus, which included an element of subjective, clinical judgment. Our SDM expert sample was small and limited to authors of the papers we reviewed, potentially restricting the range of viewpoints received. Lastly, the SDM 3 Circle Model highlights key concept areas rather than all possible factors that influence SDM.

CONCLUSIONS

We present a peer-reviewed, literature-based SDM model capable of accounting for the unique circumstances and challenges of SDM in the hospital. The SDM 3 Circle Model identifies the primary categories of variables thought to influence SDM, places them in a shared environmental frame, and visually represents their interactive nature. A multistep representation of the SDM process further illustrates how the unique features and challenges of hospitalization might exert influence at various points as patients and providers reach a shared decision. As the interrelationships of patient and provider/team, medical context, and the environmental frame in which they occur are better understood, more effective and targeted interventions to promote SDM can be developed and evaluated.

Acknowledgments

The authors would like to thank Evans Whitaker for his assistance with the literature review and the Patient Engagement Project volunteers for their support and assistance with data collection.

Disclosure

Financial support for this study was provided entirely by a grant from NIH/NCCIH (grant #R25 AT006573, awarded to Dr. Jason Satterfield). The funding agreement ensured the authors’ independence in designing the study, interpreting the data, writing, and publishing the report. The following authors are employed by the sponsor: Stephanie Rennke, MD, Patrick Yuan, BA, Brad Monash, MD, Rebecca Blankenburg, MD, MPH, Ian Chua, MD, Stephanie Harman, MD, Debbie S. Sakai, MD, Joan F. Hilton, DSc, MPH., and Jason Satterfield, PhD.

 

Venous thromboembolism (VTE) affects more than 1 million patients and costs the US healthcare system more than $1.5 billion annually.¹ Inherited and acquired thrombophilias have been perceived as important risk factors in assessing the risk of VTE recurrence and guiding the duration of anticoagulation.

Thrombophilias increase the risk of a first thrombotic event, but existing data have failed to demonstrate the usefulness of routine thrombophilia screening on subsequent management.^2,3 Moreover, thrombophilia testing ordered in the context of an inpatient hospitalization is limited by confounding factors, especially during an acute thrombotic event or in the setting of concurrent anticoagulation.⁴

Recognizing the costliness of routine thrombophilia testing, The American Society of Hematology introduced its Choosing Wisely campaign in 2013 in an effort to reduce test ordering in the setting of provoked VTEs with a major transient risk factor.⁵ In order to define current practice behavior at our institution, we conducted a retrospective study to determine the magnitude and financial impact of potentially inappropriate thrombophilia testing in the inpatient setting.

METHODS

We performed a retrospective analysis of thrombophilia testing across all inpatient services at a large, quaternary-care academic institution over a 2-year period. Electronic medical record data containing all thrombophilia tests ordered on inpatients from June 2013 to June 2015 were obtained. This study was exempt from institutional review board approval.

Inclusion criteria included any inpatient for which thrombophilia testing occurred. Patients were excluded if testing was ordered in the absence of VTE or arterial thrombosis or if it was ordered as part of a work-up for another medical condition (see Supplementary Material).

Thrombophilia testing was defined as any of the following: inherited thrombophilias (Factor V Leiden or prothrombin 20210 gene mutations, antithrombin, or protein C or S activity levels) or acquired thrombophilias (lupus anticoagulant [Testing refers to the activated partial thromboplastin time lupus assay.], beta-2 glycoprotein 1 immunoglobulins M and G, anticardiolipin immunoglobulins M and G, dilute Russell’s viper venom time, or JAK2 V617F mutations).

Extracted data included patient age, sex, type of thrombophilia test ordered, ordering primary service, admission diagnosis, and objective confirmation of thrombotic events. The indication for test ordering was determined via medical record review of the patient’s corresponding hospitalization. Each test was evaluated in the context of the patient’s presenting history, hospital course, active medications, accompanying laboratory and radiographic studies, and consultant recommendations to arrive at a conclusion regarding both the test’s reason for ordering and whether its indication was “inappropriate,” “appropriate,” or “equivocal.” Cost data were obtained through the Centers for Medicare & Medicaid Services (CMS) Clinical Laboratory Fee Schedule for 2016 (see Supplementary Material).⁶

The criteria for defining test appropriateness were formulated by utilizing a combination of major society guidelines and literature review.^5,7-10 The criteria placed emphasis upon the ordered tests’ clinical relevance and reliability and were subsequently reviewed by a senior hematologist with specific expertise in thrombosis (see Supplementary Material).

Two internal medicine resident physician data reviewers independently evaluated the ordered tests. To ensure consistency between reviewers, a sample of identical test orders was compared for concordance, and a Cohen’s kappa coefficient was calculated. For purposes of analysis, equivocal orders were included under the appropriate category, as this study focused on the quantification of potentially inappropriate ordering practices. Pearson chi-square testing was performed in order to compare ordering practices between services using Stata.¹¹

RESULTS

In total, we reviewed 2179 individual tests, of which 362 (16.6%) were excluded. The remaining 1817 tests involved 299 patients across 26 primary specialties. Fifty-two (2.9% of orders) were ultimately deemed equivocal. The Table illustrates the overall proportion and cost of inappropriate test ordering as well as testing characteristics of the most commonly encountered thrombotic diagnoses. The Figure illustrates the proportion of potentially inappropriate test ordering with its associated cost by test type.

Orders for Factor V Leiden, prothrombin 20210, and protein C and S activity levels were most commonly deemed inappropriate due to the test results’ failure to alter clinical management (97.3%, 99.2%, 99.4% of their inappropriate orders, respectively). Antithrombin testing (59.4%) was deemed inappropriate most commonly in the setting of acute thrombosis. The lupus anticoagulant (82.8%) was inappropriately ordered most frequently in the setting of concurrent anticoagulation.

Ordering practices were then compared between nonteaching and teaching inpatient general medicine services. We observed a higher proportion of inappropriate tests ordered by the nonteaching services as compared to the teaching services (120 of 173 orders [69.4%] versus 125 of 320 [39.1%], respectively; P < 0.001).

The interreviewer kappa coefficient was 0.82 (P < 0.0001).

 

DISCUSSION

This retrospective analysis represents one of the largest examinations of inpatient thrombophilia testing practices to date. Our results illustrate the high prevalence and significant financial impact of potentially inappropriate thrombophilia testing conducted in the inpatient setting. The data confirm that, per our defined criteria, more than 90% of inherited thrombophilia testing was potentially inappropriate while the majority of acquired thrombophilia testing was appropriate, with the exception of the lupus anticoagulant.

Even when appropriately ordered, studies suggest that positive thrombophilia screening results fail to impact outcomes in most patients with VTE. In an effort to evaluate positive results’ potential to provide a basis from which to extend the duration of anticoagulation, and therefore reduce the risk of a recurrent VTE, a case-control analysis was performed on a series of patients with a first-VTE event (Multiple Environmental and Genetic Assessment of risk factors for venous thrombosis [MEGA] study).³ In examining the odds ratio (OR) for recurrence between patients who did or did not undergo testing for Factor V Leiden, antithrombin, or protein C or S activity, the data failed to show an impact of testing on the risk of VTE recurrence (OR 1.2; confidence interval, 0.8-1.8). In fact, decision making has increasingly relied on patients’ clinical characteristics rather than thrombophilia test results to guide anticoagulation duration after incident VTEs. A 2017 study illustrated that when using a clinical decision rule (Clinical Decision Rule Validation Study to Predict Low Recurrent Risk in Patients With Unprovoked Venous Thromboembolism [REVERSE criteria]) in patients with a first, unprovoked VTE, routine thrombophilia screening added little to determining the need for prolonged anticoagulation.¹² These findings support the limited clinical utility of current test ordering practices for the prediction and management of recurrent venous thrombosis.

Regarding the acquired thrombophilias, antiphospholipid antibody testing was predominantly ordered in a justified manner, which is consistent with the notion that test results could affect clinical management, such as anticoagulation duration or choice of anticoagulant.¹³ However, the validity of lupus anticoagulant testing was limited by the frequency of patients on concurrent anticoagulation.

Financially, the cumulative cost associated with inappropriate ordering was substantial, regardless of the thrombotic event in question. Moreover, our calculated costs are derived from CMS reimbursement rates and likely underestimate the true financial impact of errant testing given that commercial laboratories frequently charge at rates several-fold higher. On a national scale, prior analyses have suggested that the annual cost of thrombophilia testing, based on typical commercial rates, ranges from $300 million to $672 million.¹⁴

Researchers in prior studies have similarly examined the frequency of inappropriate thrombophilia testing and methods to reduce it. Researchers in a 2014 study demonstrated initially high rates of inappropriate inherited thrombophilia testing, and then showed marked reductions in testing and cost savings across multiple specialties following the introduction of a flowchart on a preprinted order form.¹⁵ Our findings provide motivation to perform similar endeavors.

The proportional difference of inappropriate ordering observed between nonteaching- and teaching-medicine services indicates a potential role for educational interventions. We recently completed a series of lectures on high-value thrombophilia ordering for residents and are actively analyzing its impact on subsequent ordering practices. We are also piloting an electronic best practice advisory for thrombophilia test ordering. Though the advisory may be overridden, providers are asked to provide justification for doing so on a voluntary basis. We plan to evaluate its effect on our findings reported in this study.

We acknowledge that our exclusion criteria resulted in the omission of testing across a spectrum of nonthrombotic clinical conditions, raising the question of selection bias. Because there are no established guidelines to determine the appropriateness of testing in these scenarios, we chose to limit the analysis of errant ordering to the context of thrombotic events. Other limitations of this study include the analysis of equivocal orders as appropriate. However, because equivocal ordering represented less than 3% of all analyzed orders, including these as inappropriate would not have significantly altered our findings.

CONCLUSIONS

A review of thrombophilia testing practices at our institution demonstrated that inappropriate testing in the inpatient setting is a frequent phenomenon associated with a significant financial impact. This effect was more pronounced in inherited versus acquired thrombophilia testing. Testing was frequently confounded and often failed to impact patients’ short- or long-term clinical management, regardless of the result.

These findings serve as a strong impetus to reduce the burden of routine thrombophilia testing during hospital admissions. Our data demonstrate a need for institution-wide changes such as implementing best practice advisories, introducing ordering restrictions, and conducting educational interventions in order to reduce unnecessary expenditures and improve patient care.

 

Disclosure

The authors have nothing to disclose.

Venous thromboembolism (VTE) affects more than 1 million patients and costs the US healthcare system more than $1.5 billion annually.¹ Inherited and acquired thrombophilias have been perceived as important risk factors in assessing the risk of VTE recurrence and guiding the duration of anticoagulation.

Thrombophilias increase the risk of a first thrombotic event, but existing data have failed to demonstrate the usefulness of routine thrombophilia screening on subsequent management.^2,3 Moreover, thrombophilia testing ordered in the context of an inpatient hospitalization is limited by confounding factors, especially during an acute thrombotic event or in the setting of concurrent anticoagulation.⁴

Recognizing the costliness of routine thrombophilia testing, The American Society of Hematology introduced its Choosing Wisely campaign in 2013 in an effort to reduce test ordering in the setting of provoked VTEs with a major transient risk factor.⁵ In order to define current practice behavior at our institution, we conducted a retrospective study to determine the magnitude and financial impact of potentially inappropriate thrombophilia testing in the inpatient setting.

METHODS

We performed a retrospective analysis of thrombophilia testing across all inpatient services at a large, quaternary-care academic institution over a 2-year period. Electronic medical record data containing all thrombophilia tests ordered on inpatients from June 2013 to June 2015 were obtained. This study was exempt from institutional review board approval.

Inclusion criteria included any inpatient for which thrombophilia testing occurred. Patients were excluded if testing was ordered in the absence of VTE or arterial thrombosis or if it was ordered as part of a work-up for another medical condition (see Supplementary Material).

Thrombophilia testing was defined as any of the following: inherited thrombophilias (Factor V Leiden or prothrombin 20210 gene mutations, antithrombin, or protein C or S activity levels) or acquired thrombophilias (lupus anticoagulant [Testing refers to the activated partial thromboplastin time lupus assay.], beta-2 glycoprotein 1 immunoglobulins M and G, anticardiolipin immunoglobulins M and G, dilute Russell’s viper venom time, or JAK2 V617F mutations).

Extracted data included patient age, sex, type of thrombophilia test ordered, ordering primary service, admission diagnosis, and objective confirmation of thrombotic events. The indication for test ordering was determined via medical record review of the patient’s corresponding hospitalization. Each test was evaluated in the context of the patient’s presenting history, hospital course, active medications, accompanying laboratory and radiographic studies, and consultant recommendations to arrive at a conclusion regarding both the test’s reason for ordering and whether its indication was “inappropriate,” “appropriate,” or “equivocal.” Cost data were obtained through the Centers for Medicare & Medicaid Services (CMS) Clinical Laboratory Fee Schedule for 2016 (see Supplementary Material).⁶

The criteria for defining test appropriateness were formulated by utilizing a combination of major society guidelines and literature review.^5,7-10 The criteria placed emphasis upon the ordered tests’ clinical relevance and reliability and were subsequently reviewed by a senior hematologist with specific expertise in thrombosis (see Supplementary Material).

Two internal medicine resident physician data reviewers independently evaluated the ordered tests. To ensure consistency between reviewers, a sample of identical test orders was compared for concordance, and a Cohen’s kappa coefficient was calculated. For purposes of analysis, equivocal orders were included under the appropriate category, as this study focused on the quantification of potentially inappropriate ordering practices. Pearson chi-square testing was performed in order to compare ordering practices between services using Stata.¹¹

RESULTS

In total, we reviewed 2179 individual tests, of which 362 (16.6%) were excluded. The remaining 1817 tests involved 299 patients across 26 primary specialties. Fifty-two (2.9% of orders) were ultimately deemed equivocal. The Table illustrates the overall proportion and cost of inappropriate test ordering as well as testing characteristics of the most commonly encountered thrombotic diagnoses. The Figure illustrates the proportion of potentially inappropriate test ordering with its associated cost by test type.

Orders for Factor V Leiden, prothrombin 20210, and protein C and S activity levels were most commonly deemed inappropriate due to the test results’ failure to alter clinical management (97.3%, 99.2%, 99.4% of their inappropriate orders, respectively). Antithrombin testing (59.4%) was deemed inappropriate most commonly in the setting of acute thrombosis. The lupus anticoagulant (82.8%) was inappropriately ordered most frequently in the setting of concurrent anticoagulation.

Ordering practices were then compared between nonteaching and teaching inpatient general medicine services. We observed a higher proportion of inappropriate tests ordered by the nonteaching services as compared to the teaching services (120 of 173 orders [69.4%] versus 125 of 320 [39.1%], respectively; P < 0.001).

The interreviewer kappa coefficient was 0.82 (P < 0.0001).

 

DISCUSSION

This retrospective analysis represents one of the largest examinations of inpatient thrombophilia testing practices to date. Our results illustrate the high prevalence and significant financial impact of potentially inappropriate thrombophilia testing conducted in the inpatient setting. The data confirm that, per our defined criteria, more than 90% of inherited thrombophilia testing was potentially inappropriate while the majority of acquired thrombophilia testing was appropriate, with the exception of the lupus anticoagulant.

Even when appropriately ordered, studies suggest that positive thrombophilia screening results fail to impact outcomes in most patients with VTE. In an effort to evaluate positive results’ potential to provide a basis from which to extend the duration of anticoagulation, and therefore reduce the risk of a recurrent VTE, a case-control analysis was performed on a series of patients with a first-VTE event (Multiple Environmental and Genetic Assessment of risk factors for venous thrombosis [MEGA] study).³ In examining the odds ratio (OR) for recurrence between patients who did or did not undergo testing for Factor V Leiden, antithrombin, or protein C or S activity, the data failed to show an impact of testing on the risk of VTE recurrence (OR 1.2; confidence interval, 0.8-1.8). In fact, decision making has increasingly relied on patients’ clinical characteristics rather than thrombophilia test results to guide anticoagulation duration after incident VTEs. A 2017 study illustrated that when using a clinical decision rule (Clinical Decision Rule Validation Study to Predict Low Recurrent Risk in Patients With Unprovoked Venous Thromboembolism [REVERSE criteria]) in patients with a first, unprovoked VTE, routine thrombophilia screening added little to determining the need for prolonged anticoagulation.¹² These findings support the limited clinical utility of current test ordering practices for the prediction and management of recurrent venous thrombosis.

Regarding the acquired thrombophilias, antiphospholipid antibody testing was predominantly ordered in a justified manner, which is consistent with the notion that test results could affect clinical management, such as anticoagulation duration or choice of anticoagulant.¹³ However, the validity of lupus anticoagulant testing was limited by the frequency of patients on concurrent anticoagulation.

Financially, the cumulative cost associated with inappropriate ordering was substantial, regardless of the thrombotic event in question. Moreover, our calculated costs are derived from CMS reimbursement rates and likely underestimate the true financial impact of errant testing given that commercial laboratories frequently charge at rates several-fold higher. On a national scale, prior analyses have suggested that the annual cost of thrombophilia testing, based on typical commercial rates, ranges from $300 million to $672 million.¹⁴

Researchers in prior studies have similarly examined the frequency of inappropriate thrombophilia testing and methods to reduce it. Researchers in a 2014 study demonstrated initially high rates of inappropriate inherited thrombophilia testing, and then showed marked reductions in testing and cost savings across multiple specialties following the introduction of a flowchart on a preprinted order form.¹⁵ Our findings provide motivation to perform similar endeavors.

The proportional difference of inappropriate ordering observed between nonteaching- and teaching-medicine services indicates a potential role for educational interventions. We recently completed a series of lectures on high-value thrombophilia ordering for residents and are actively analyzing its impact on subsequent ordering practices. We are also piloting an electronic best practice advisory for thrombophilia test ordering. Though the advisory may be overridden, providers are asked to provide justification for doing so on a voluntary basis. We plan to evaluate its effect on our findings reported in this study.

We acknowledge that our exclusion criteria resulted in the omission of testing across a spectrum of nonthrombotic clinical conditions, raising the question of selection bias. Because there are no established guidelines to determine the appropriateness of testing in these scenarios, we chose to limit the analysis of errant ordering to the context of thrombotic events. Other limitations of this study include the analysis of equivocal orders as appropriate. However, because equivocal ordering represented less than 3% of all analyzed orders, including these as inappropriate would not have significantly altered our findings.

CONCLUSIONS

A review of thrombophilia testing practices at our institution demonstrated that inappropriate testing in the inpatient setting is a frequent phenomenon associated with a significant financial impact. This effect was more pronounced in inherited versus acquired thrombophilia testing. Testing was frequently confounded and often failed to impact patients’ short- or long-term clinical management, regardless of the result.

These findings serve as a strong impetus to reduce the burden of routine thrombophilia testing during hospital admissions. Our data demonstrate a need for institution-wide changes such as implementing best practice advisories, introducing ordering restrictions, and conducting educational interventions in order to reduce unnecessary expenditures and improve patient care.

 

Disclosure

The authors have nothing to disclose.

Venous thromboembolism (VTE) affects more than 1 million patients and costs the US healthcare system more than $1.5 billion annually.¹ Inherited and acquired thrombophilias have been perceived as important risk factors in assessing the risk of VTE recurrence and guiding the duration of anticoagulation.

Thrombophilias increase the risk of a first thrombotic event, but existing data have failed to demonstrate the usefulness of routine thrombophilia screening on subsequent management.^2,3 Moreover, thrombophilia testing ordered in the context of an inpatient hospitalization is limited by confounding factors, especially during an acute thrombotic event or in the setting of concurrent anticoagulation.⁴

Recognizing the costliness of routine thrombophilia testing, The American Society of Hematology introduced its Choosing Wisely campaign in 2013 in an effort to reduce test ordering in the setting of provoked VTEs with a major transient risk factor.⁵ In order to define current practice behavior at our institution, we conducted a retrospective study to determine the magnitude and financial impact of potentially inappropriate thrombophilia testing in the inpatient setting.

METHODS

We performed a retrospective analysis of thrombophilia testing across all inpatient services at a large, quaternary-care academic institution over a 2-year period. Electronic medical record data containing all thrombophilia tests ordered on inpatients from June 2013 to June 2015 were obtained. This study was exempt from institutional review board approval.

Inclusion criteria included any inpatient for which thrombophilia testing occurred. Patients were excluded if testing was ordered in the absence of VTE or arterial thrombosis or if it was ordered as part of a work-up for another medical condition (see Supplementary Material).

Thrombophilia testing was defined as any of the following: inherited thrombophilias (Factor V Leiden or prothrombin 20210 gene mutations, antithrombin, or protein C or S activity levels) or acquired thrombophilias (lupus anticoagulant [Testing refers to the activated partial thromboplastin time lupus assay.], beta-2 glycoprotein 1 immunoglobulins M and G, anticardiolipin immunoglobulins M and G, dilute Russell’s viper venom time, or JAK2 V617F mutations).

Extracted data included patient age, sex, type of thrombophilia test ordered, ordering primary service, admission diagnosis, and objective confirmation of thrombotic events. The indication for test ordering was determined via medical record review of the patient’s corresponding hospitalization. Each test was evaluated in the context of the patient’s presenting history, hospital course, active medications, accompanying laboratory and radiographic studies, and consultant recommendations to arrive at a conclusion regarding both the test’s reason for ordering and whether its indication was “inappropriate,” “appropriate,” or “equivocal.” Cost data were obtained through the Centers for Medicare & Medicaid Services (CMS) Clinical Laboratory Fee Schedule for 2016 (see Supplementary Material).⁶

The criteria for defining test appropriateness were formulated by utilizing a combination of major society guidelines and literature review.^5,7-10 The criteria placed emphasis upon the ordered tests’ clinical relevance and reliability and were subsequently reviewed by a senior hematologist with specific expertise in thrombosis (see Supplementary Material).

Two internal medicine resident physician data reviewers independently evaluated the ordered tests. To ensure consistency between reviewers, a sample of identical test orders was compared for concordance, and a Cohen’s kappa coefficient was calculated. For purposes of analysis, equivocal orders were included under the appropriate category, as this study focused on the quantification of potentially inappropriate ordering practices. Pearson chi-square testing was performed in order to compare ordering practices between services using Stata.¹¹

RESULTS

In total, we reviewed 2179 individual tests, of which 362 (16.6%) were excluded. The remaining 1817 tests involved 299 patients across 26 primary specialties. Fifty-two (2.9% of orders) were ultimately deemed equivocal. The Table illustrates the overall proportion and cost of inappropriate test ordering as well as testing characteristics of the most commonly encountered thrombotic diagnoses. The Figure illustrates the proportion of potentially inappropriate test ordering with its associated cost by test type.

Orders for Factor V Leiden, prothrombin 20210, and protein C and S activity levels were most commonly deemed inappropriate due to the test results’ failure to alter clinical management (97.3%, 99.2%, 99.4% of their inappropriate orders, respectively). Antithrombin testing (59.4%) was deemed inappropriate most commonly in the setting of acute thrombosis. The lupus anticoagulant (82.8%) was inappropriately ordered most frequently in the setting of concurrent anticoagulation.

Ordering practices were then compared between nonteaching and teaching inpatient general medicine services. We observed a higher proportion of inappropriate tests ordered by the nonteaching services as compared to the teaching services (120 of 173 orders [69.4%] versus 125 of 320 [39.1%], respectively; P < 0.001).

The interreviewer kappa coefficient was 0.82 (P < 0.0001).

 

DISCUSSION

This retrospective analysis represents one of the largest examinations of inpatient thrombophilia testing practices to date. Our results illustrate the high prevalence and significant financial impact of potentially inappropriate thrombophilia testing conducted in the inpatient setting. The data confirm that, per our defined criteria, more than 90% of inherited thrombophilia testing was potentially inappropriate while the majority of acquired thrombophilia testing was appropriate, with the exception of the lupus anticoagulant.

Even when appropriately ordered, studies suggest that positive thrombophilia screening results fail to impact outcomes in most patients with VTE. In an effort to evaluate positive results’ potential to provide a basis from which to extend the duration of anticoagulation, and therefore reduce the risk of a recurrent VTE, a case-control analysis was performed on a series of patients with a first-VTE event (Multiple Environmental and Genetic Assessment of risk factors for venous thrombosis [MEGA] study).³ In examining the odds ratio (OR) for recurrence between patients who did or did not undergo testing for Factor V Leiden, antithrombin, or protein C or S activity, the data failed to show an impact of testing on the risk of VTE recurrence (OR 1.2; confidence interval, 0.8-1.8). In fact, decision making has increasingly relied on patients’ clinical characteristics rather than thrombophilia test results to guide anticoagulation duration after incident VTEs. A 2017 study illustrated that when using a clinical decision rule (Clinical Decision Rule Validation Study to Predict Low Recurrent Risk in Patients With Unprovoked Venous Thromboembolism [REVERSE criteria]) in patients with a first, unprovoked VTE, routine thrombophilia screening added little to determining the need for prolonged anticoagulation.¹² These findings support the limited clinical utility of current test ordering practices for the prediction and management of recurrent venous thrombosis.

Regarding the acquired thrombophilias, antiphospholipid antibody testing was predominantly ordered in a justified manner, which is consistent with the notion that test results could affect clinical management, such as anticoagulation duration or choice of anticoagulant.¹³ However, the validity of lupus anticoagulant testing was limited by the frequency of patients on concurrent anticoagulation.

Financially, the cumulative cost associated with inappropriate ordering was substantial, regardless of the thrombotic event in question. Moreover, our calculated costs are derived from CMS reimbursement rates and likely underestimate the true financial impact of errant testing given that commercial laboratories frequently charge at rates several-fold higher. On a national scale, prior analyses have suggested that the annual cost of thrombophilia testing, based on typical commercial rates, ranges from $300 million to $672 million.¹⁴

Researchers in prior studies have similarly examined the frequency of inappropriate thrombophilia testing and methods to reduce it. Researchers in a 2014 study demonstrated initially high rates of inappropriate inherited thrombophilia testing, and then showed marked reductions in testing and cost savings across multiple specialties following the introduction of a flowchart on a preprinted order form.¹⁵ Our findings provide motivation to perform similar endeavors.

The proportional difference of inappropriate ordering observed between nonteaching- and teaching-medicine services indicates a potential role for educational interventions. We recently completed a series of lectures on high-value thrombophilia ordering for residents and are actively analyzing its impact on subsequent ordering practices. We are also piloting an electronic best practice advisory for thrombophilia test ordering. Though the advisory may be overridden, providers are asked to provide justification for doing so on a voluntary basis. We plan to evaluate its effect on our findings reported in this study.

We acknowledge that our exclusion criteria resulted in the omission of testing across a spectrum of nonthrombotic clinical conditions, raising the question of selection bias. Because there are no established guidelines to determine the appropriateness of testing in these scenarios, we chose to limit the analysis of errant ordering to the context of thrombotic events. Other limitations of this study include the analysis of equivocal orders as appropriate. However, because equivocal ordering represented less than 3% of all analyzed orders, including these as inappropriate would not have significantly altered our findings.

CONCLUSIONS

A review of thrombophilia testing practices at our institution demonstrated that inappropriate testing in the inpatient setting is a frequent phenomenon associated with a significant financial impact. This effect was more pronounced in inherited versus acquired thrombophilia testing. Testing was frequently confounded and often failed to impact patients’ short- or long-term clinical management, regardless of the result.

These findings serve as a strong impetus to reduce the burden of routine thrombophilia testing during hospital admissions. Our data demonstrate a need for institution-wide changes such as implementing best practice advisories, introducing ordering restrictions, and conducting educational interventions in order to reduce unnecessary expenditures and improve patient care.

 

Disclosure

The authors have nothing to disclose.