Clinical Decision Science: Proof of concept

Clinical decision science, a newly identified area of scholarship, describes how clinical research is used for a patient, within the context of their unique social conditions. We hypothesize that physicians use sociocultural context as an important input to their decision making. We performed a prospective, randomized, double-blind mixed methods study. Family medicine faculty and residents at a community hospital family medicine residency were included in the study. After academic journal club discussing a primary research paper, physicians were asked if they would prescribe medication for a patient who was similar to the subjects in a research paper. However, social and cultural context was given to the cases. The physicians were block randomized into two groups; Group A was given a case with a patient who had a social and cultural context more conducive to lifestyle modification, while group B had a case more conducive to prescribing medication. Primary outcome was whether or not physicians prescribed medication, allowing for a 2x2 table for analysis. We also asked a free response question regarding the reasoning for their decision and performed qualitative analysis. In group A (n=14), no subjects prescribed medication. In group B (n=18), six subjects prescribed medication. (chi-square statistic with Yates correction 9.95, p=0.001). Thematic analysis in group A showed 22 statements related to medical practice or disease characteristics, and 25 statements related to familial relationships and patient preferences. In group B, subjects who prescribed medication made 30 statements related to medical practice, and 13 statements related to patient preference. Subjects who did not prescribe medication made 10 statements related to medical practice, and 1 statement related to patient preference. This study demonstrates social context of individual patients, together with evidence, affects clinical decision and management. We propose further study into how this affects physician decision making, a scholarly field we call Clinical Decision Science.


Introduction
Despite the inherent complexity of clinical decision making, the process by which clinicians make decisions has only been studied on a sociologic level. Sociological research can only describe behavior of groups of physicians and cannot inform how an individual clinician generates a clinical decision. 1,2 Experimental research that takes patient, physician, and practice setting characteristics into account and is related to clinical problems encountered in day-to day practice does not exist. 3,4 Clark et. al. state "...focus on physicians' thinking has been accompanied by rather little attention to social context..." 4 Similarly, Bachur et. al. in a more recent paper note, "...literature that is focused on clinical practice...is almost nonexistent." 5 The ideal of a biopsychosocial model of clinical practice has been proposed. 6,7 Although it may be achieved in clinical practice, it has never been studied in the medical literature. Although doctors excel in the biological domain and have some skills in the psychological domain, the role that a patient's social context plays in clinical practice has largely been ignored. This is understandable because physicians receive no formal training to assess social context. 8  This research report is the first randomized controlled trial demonstrating that social context changes clinical decisions. Our hypothesis was that physicians' prescribing behavior is influenced by social context of the patient.

Methods
This was a prospective, randomized, double-blind, mixed methods study. We obtained IRB approval.
Prior to initiating the study, a power analysis was done. We purposefully created extremes in descriptors of the patients' social context to achieve the greatest effect size possible between two standardized clinical scenarios. This was in keeping with the "proof of concept" purpose of the research and the limitation we had on total number of possible cases. Using an alpha of 0.05, beta of 0.2 (Power of 80% to detect a difference) and a 50% effect size between the two groups, the study required 13 participants in each group.
The study was performed in a family medicine residency at a community hospital. Participants included faculty and residents who attend journal club. A list of all faculty and residents was obtained from the program coordinator. We excluded faculty and residents who were study investigators.
At our institution, Journal Club is focused on learning to critically appraise clinical research. For this study, the paper that was critically appraised and the surveys about clinical decisions were thematically linked-meeting the standard of "Clinical Research in Practice". The critically appraised clinical research paper compared lifestyle modification versus metformin for patients with impaired fasting glucose. 11 For a brief description of the Journal Club paper and the salient points discussed, see Appendix A.
Prior to journal club, study materials were distributed to all potential participants in a sealed envelope. Potential participants were instructed not to open the envelope until after journal club.
Verbal recruitment and consent using an IRB-approved Information Sheet were explained by one of the investigators (AN). Participants were stratified by year of practice and then randomized into two groups with a block four design. Each of the two groups was assigned a standardized, written patient case. Both standardized cases had identical biomedical facts, but markedly different social contexts. The contrasting social contexts used the parameters of familial support, patient preferences, emotions, institutional constraints, and patient reported conversations. 12 In group A, the social context favored lifestyle modification. In group B, the social context favored pharmacotherapy.
The social context of the patient in the standardized case comprised the independent variable. Included with the patient case was a survey, asking if the participant would prescribe medication (metformin) to the patient depicted in the standardized clinical scenario and additionally to describe their medical reasoning for that decision. The decision to prescribe or not prescribe medication was the dependent variable. The clinical scenarios and surveys are represented in Appendices B & C.
Immediately after the journal club discussion, participants opened the envelopes containing the research information sheet, written standardized patient case details, and the data collection survey. Participants were given ten minutes to read the materials and complete the survey. After finishing, participants placed their survey in an inner envelope marked, "ANONYMOUS & CONFIDENTIAL" and then sealed the inner envelope themselves. Participants then placed the inner envelope into another sealed envelope. If attendees wished to decline participation anonymously, they were instructed to place a blank survey in the inner envelope. The outer sealed envelopes were labelled with the participants' name, so that a response rate could be calculated. However, the inner envelope and survey, did not contain any identifying marks or information and participants were instructed not to put identifying information on any research material. The envelopes were collected by investigators, who calculated the response rate. The envelopes were then given to a trusted agent. The trusted agent process was requested by the IRB and was a person that has no influence over respondents; this was to minimize coercion because the respondents included trainees. The trusted agent opened the outer envelopes and removed the anonymous and confidential inner envelopes, and immediately mixed them to maintain anonymity and confidentiality. In this way, no individual survey could be linked to any individual participant or attendee, whether they participated or declined participation.
The inner envelopes were then opened, and data was collected and analyzed. The decision to prescribe medication was recorded as a dichotomous dependent variable, allowing a 2x2 table, comparing social context to medication prescription behavior. The free response section asking participants to explain their decision was analyzed using content analysis and thematic analysis. All the investigators participated in the analysis and developed a consensus when applying labels to the qualitive data. Content analysis was done by pile sorting identical or nearly identical responses. Thematic analysis was done by applying a label describing a collection of related content items.

Results
A total of 38 potential participants were identified. We received 33 responses, and 1 response was blank indicating nonparticipation. In total, we received 32 completed surveys, for a response rate of 84%.
In group A (social context favoring lifestyle intervention), 0/14 subjects prescribed metformin to the patient. In group B (social context favoring medication), 12/18 subjects prescribed metformin to the patient. Using Fisher's exact test, this was statistically significant with p=0.0001. (Chi-square statistic with Yates correction 9.95, p=0.001) Figure 1 shows the labels applied by the investigators summarizing the qualitative responses explaining participant prescribing behavior. Many respondents recorded more than one rationale. Table 1 gives the specific content with frequencies in Group A. Table  2 gives the specific content with frequencies in Group B.   He has tried to make changes many times.
Thematic analysis in group A showed 22 statements related to medical practice or disease characteristics, and 25 statements related to familial relationships and patient preferences. In group B, subjects who prescribed medication made 30 statements related to medical practice, and 13 statements related to patient preference. Subjects who did not prescribe medication made 10 statements related to medical practice, and 1 statement related to patient preference. Group B was the only group where a prescription was given-one of the noteworthy findings of this study.

Results
Although this is a small pilot study, our data document a "proof of concept" that physicians include social context of the patient when making clinical decisions. This finding is a core tenet of Clinical Decision Science. We perceive an educational focus on biomedical facts and algorithms. However, the process of clinical decision making is complex and influenced by each patient's social context. Unfortunately, clinicians are not explicitly taught how to do this, and they must learn by observation in an apprenticeship learning environment. Academic focus on clinical decision science can expedite learning this important aspect of clinical practice.