Translating clinical and patient-reported data to tailored shared decision reports with predictive analytics for knee and hip arthritis

The A.S.K. web-based informatics system includes the following functions in order to generate a tailored report at the point of clinical decision-making: (1) email delivery of unique web links to collect PRO and patient-reported clinical risk data (e.g., smoking, BMI, depressive symptoms) from patients at home or in office prior to clinician visit; (2) automated statistical processing of patient responses by comparing each individual to the FORCE-TJR national registry of 35,000 TJR patients with parallel demographic, comorbidity, and PRO data; and (3) computation of individualized estimates of likely pain relief and functional gain after surgery. The A.S.K. system then generates a real-time report with (a) the individual’s pain and function scores compared to national data, (b) key comorbid risk factors, (c) a decision grid for non-surgical options, and (d) individualized estimates of likely pain relief and functional gain based on the experience of similar patients to support shared decision-making between the patient and clinician at the office visit. Finally, (4) patient data are returned through batch feeds to the local sites’ EHR systems. Figure 1 illustrates the steps to produce the A.S.K. report for office use and to store data both in the local EHR and the national registry.

Increasing numbers of arthroplasty clinical practices use PROs to measure the severity of knee and hip osteoarthritis patients’ pain and function [10, 11]. Prior research demonstrates that PROs, combined with demographic and clinical risk factors, can explain statistical variation in 30 day readmissions and in pain and function one year after TJR [12‐14]. Thus, the A.S.K. report wanted to display pain and function sub-scores of the Knee and Hip Osteoarthritis Outcome Scores (KOOS, HOOS) [15, 16] as well as patient risk factors that research associates with post-TJR symptom improvement or complications. Predictive analytics could then use these data to generate likely TJR outcomes based on patients with similar clinical symptoms and PROs captured within the national FORCE-TJR registry.

Our prior analyses of national data with outcomes at 12 months after TJR surgery documented wide variation in pain and function and the ability to predict discrete sub-groups of greater and lesser post-operative improvement [13]. These initial nomothetic models and subsequent FORCE-TJR analyses support that, in general, key pre-operative patient symptom and clinical variables are associated with greater or lesser pain relief and functional gain after surgery [17‐19]. In addition, preliminary multivariable models developed in preparation for the A.S.K. study identified that pre-TKR attributes are significantly (p < 0.05) associated with specific activities, such as a greater likelihood of pain-free walking at 12 months after surgery: not obese, no knee pain at rest, no pain walking, excellent or very good general health, not feeling down-hearted or blue, and no history of prior ipsilateral knee procedure. Similarly, models combining pre-surgical comorbidity measures and patient-reported symptoms could predict readmission or not with good discrimination (C = 0.79 (knees); 0.86 (hips)) [12]. Using parallel pre-surgical patient-reported and clinical variables, models predicting likely 12 month post-TJR joint pain, joint-specific function, and global function were developed for the A.S.K. report.

To define the content, sequence, and visual presentation of these data in the A.S.K. report, iterative clinician key informant interviews were conducted. Five experienced arthroplasty surgeons from MA, CT, NY, and PA, a nurse practitioner, and a physical therapist participated in interviews. Clinicians reviewed literature summarizing the association between key patient and clinical factors and surgical outcomes, including pain relief, functional gain, and complications. Interviews began with open-ended questions as clinicians examined different graphical representations of the data. Second, the study’s patient advisers reviewed draft reports. Interviewers probed to validate key domains to include in the A.S.K. report. The interviewer captured initial impressions and observations in “quick notes”. The interview transcripts were analyzed using standard qualitative analysis techniques [20, 21] and compared with results from other studies examining the use of educational aids in orthopedics. Grounded theory coding techniques were employed [22] focusing on the underlying processes of TJR decision-making. Transcripts for each interview were reviewed and themes and categories of like responses identified (e.g., by question topic, by positive/negative). Multiple drafts of the A.S.K. report templates with color-coded symptom severity were reviewed with the clinicians to refine specifications. In addition, a draft decision grid with common non-surgical treatment options based on professional society recommendations was reviewed [23, 24].

To further refine the A.S.K. report presentation for clarity and interpretation, patients were recruited from orthopedic centers of participating surgeon advisors, with attention to diverse races and ethnicities. To anticipate the needs of future report users, adults (n = 16) who had recently made decisions about undergoing TJR surgery were identified to guide refinement of report presentation, priority outcomes, and information for non-operative treatments. Test patients included both adults who decided for and against elective TJR surgery to parallel future users of the report. The mean age of participants was 69.7 years (range 66–83 years); 56% male; 22% Hispanic and 22% Black; 22% had not elected TJR at the time of the interview. Semi-structured, in-person interviews were conducted by a single experienced clinician-qualitative researcher and were digitally recorded. In each interview, patients reviewed multiple A.S.K. report templates with optional data displays and were invited to describe their interpretation, provide feedback, and make suggestions to improve clarity. (See Fig. 2). Exploratory questions addressed: (1) ease of use of the tool; (2) comprehension of the tool content; (3) most useful elements, and most- or least-recalled elements; (4) relevance of the outcome data; and (5) general suggestions for improving the tool. Draft reports included varying visual data presentations such as bar graphs, pie graphs, word clouds, and people icons to represent proportions. Color coding with red representing the most severe symptoms and green the healthiest were applied. Reports were revised between interviews, based on sequential feedback. Consensus on report presentation was reached after nine 60- to 90-min interviews.

Using a site-level, cluster randomized design, the A.S.K. study will evaluate if this PRO-based shared decision tool enhances patients’ decision quality and/or influences treatment choices and outcomes. Because osteoarthritis is the most common disabling condition in the US and TJR is the costliest inpatient procedure, the decision about use of TJR is an important proof of concept [25]. Lessons learned from this study can inform future methods to translate large databases to individualized shared decision reports with predictive analytics.

Clinicians advised that both patient-reported symptoms (e.g., pain, limitations in function) and modifiable clinical risk factors important to predicting outcomes (e.g., BMI, smoking, emotional health) be presented simultaneously to frame both disease severity and surgical risks in the discussion. Because osteoarthritis pain can affect multiple joints, surgeons preferred that separate pain graphs be presented for right and left knees and hips. To align with emerging orthopedic norms and CMS-endorsed measures, surgeons recommended that the report first display knee and hip-specific pain scores (KOOS and HOOS, respectively) and then graph joint-specific function (KOOS or HOOS ADL) and global physical function [23, 24]. The clinical risk factors displayed are variables that are independently significant in prediction models of likely outcomes and are supported by prior research described above. Specifically, both young and old ages, diabetes [26], smoking, very high or very low BMI [27], advanced co-morbidities per Charlson index [13], co-existing pain in the non-operative knees and hips, and moderate to severe low back pain [28] have all been associated with higher complications and poorer TJR outcomes.

Clinicians endorsed the presentation of the calculated predicted likely benefits of surgery (e.g., pain relief and functional outcomes). However, because serious adverse event rates are small, and prediction models less precise, they suggested that national hospital readmission rate ranges be described [24]. Last, clinicians advised that national professional society best practice guidelines serve as the source of information for the decision grid [23]. At the conclusion of the clinician interviews, refined draft reports were developed for patients to guide the final data presentation and content.

Online Appendix 1 includes the final three-page A.S.K. report. Page 1 displays pain scores for both right and left knees and hips as well as function scores. In addition, medical and musculoskeletal risk factors associated with post-TJR outcomes are presented. Page 2 includes the predicted analytics for likely one-year post-TJR pain, function, and likelihood of being pain-free with key tasks. Of note, both arthritis-specific and generic function measures are displayed as the surgery is expected to improve joint-related function, but generic physical function, by definition, is affected by co-existing health conditions and may or may not improve. Page 3 includes a decision grid with comparisons of common non-operative treatments including medications, knee injections, and physical therapy. Patients defined the questions to include such as potential impact, length of treatment time, and need to repeat the treatment.

Both patient and clinical participants believed the content of the report was important to making a decision regarding elective TJR surgery. Participants stated that a shared decision report with individualized PROs assisted them in understanding the severity of their osteoarthritis, and informed decisions regarding elective TJR surgery. Beyond the report, a subset of participants believed that speaking with other patients who had previously made the decision for or against TJR surgery would be helpful.

The lead of the patient advisory group summarized the group’s sentiments:

In my mind [the survey] is part of the treatment plan. A lot of times they leave the patient out. They give the diagnosis and tell them to take that pill or whatever. By giving the answers to the questions, it helps you be involved in the actual decision-making and to understand why.

With PCORI funding, the A.S.K. web-based data collection and reporting system was implemented at 12 orthopedic offices with 26 participating surgeons in an ongoing site-level, cluster randomized trial. The University of Massachusetts Medical School Institutional Review Board for Human Subjects serves as the single IRB of record for all sites. In addition to IRB reviews, information system security reviews were completed at all participating sites prior to study initiation. Each site approved the A.S.K. system as meeting optimal standards for privacy and data security standards. Finally, prior to implementation, the 26 site surgeons, mid-level practitioners, and clinic staff were educated on the use of the report, the value of shared decision-making, and each of their roles in the process of data capture and report generation. Staff sessions were site specific and tailored to the local office flow.

To date, 7891 patients completed the PRO and risk assessment before the office visit. Of these, 73% had a diagnosis of knee or hip osteoarthritis making them eligible for the study and 56% of eligible patients (n = 3157) consented to complete a post-visit structured interview. Refusals to participate were uncommon (12%); “unable to reach by phone” after the office visit was the most common reason that patients were not consented. Consenting patients have a mean age of 65.7 years, 61% female, mean BMI of 32.0. See Table 1 for demographics and risk factors at time of office visit.

Offices received dedicated color printers to assure that the patient and clinician received color-coded reports for shared review during the office visit. Overall, A.S.K. reports were printed for 93% of patients with completed data. At 2 weeks post-visit, 68% recalled reviewing the report with their surgeon (range 45–81%) and a majority of patients valued the information. In addition, 86% of patients at the 6 month post-visit time point and 87% at the 12 month time point completed a repeat PRO assessment of their arthritis symptoms.

Web-based tracking reports allow sites to monitor data capture and report rates. Consented patients complete a post-visit structured interview to assess report use and decision quality. Study enrollment, shared decision report evaluation, and 6 and 12 month symptom monitoring are ongoing. The impact of components of the A.S.K. report on shared decision quality and outcomes will be analyzed according to the milestones described in the research protocol.