Clinic variation in recruitment metrics, patient characteristics and treatment use in a randomized clinical trial of osteoarthritis management

Knee and hip osteoarthritis (OA) are among the most common chronic health conditions and leading contributors to pain and disability among adults [1–6]. The prevalence of OA is on the rise, and this trend is expected to continue [7]. For example, recent data from the Framingham Osteoarthritis Study show that over the past 20 years, the prevalence of knee OA approximately tripled in men and almost doubled in women [8]. Therefore, in addition to the substantial toll of OA at the individual level [9], this health problem is a significant burden for healthcare systems [10–12].

Evidence-based guidelines emphasize that adequate management of knee and hip OA requires a combination of behavioral and clinical strategies [13–16]. However, many studies show there is underutilization of core treatment approaches [12, 17–21]. For example, although weight management and physical activity are key components of managing knee and hip OA [13, 15, 16], the majority of adults with OA are overweight and physically inactive [17, 22, 23], and provider recommendations for these behavioral strategies are infrequent [19, 24–26]. With respect to clinical care, several studies have shown low pass rates for OA quality of care indicators, including assessment of pain and function, referrals to other providers (when indicated), appropriate prescribing of pain medications, and general use of non-pharmacological treatments [20, 27–30]. These data all highlight the need for evidence-based interventions to augment the use of behavioral strategies and recommended components of clinical care for patients with knee and hip OA, with the ultimate goal of improving pain, function, and other key outcomes.

The Patient and PRovider Interventions for Managing Osteoarthritis in Primary Care (PRIMO) study is being conducted to address these gaps. Since optimal management of OA is multifactorial in nature, PRIMO is designed to simultaneously test both patient-based and provider-based interventions [31]. The provider-based intervention, which delivers patient-specific treatment recommendations to providers at the point of care, is particularly innovative; only a few studies have examined provider-based programs to enhance management of OA in clinic settings [32–34], and none have been in the U.S., to our knowledge. A particularly important feature of PRIMO is that it is being conducted in multiple primary care clinics. This is important for assessing generalizability of the interventions and for understanding whether patient or clinic characteristics may influence the feasibility and effectiveness in real-world practice.

In this manuscript we focus on describing the variation in patient and practice characteristics across the ten clinics included in PRIMO. There are four specific objectives. First, we report on practice-related and geographic variables across study clinics. Second, we present recruitment data for each study site. Third, we report on participant demographic and clinical characteristics across study sites. Fourth, we describe the use of various OA treatments at baseline across the study sites. Understanding how OA treatment use may differ across sites is important for the proposed study because the provider-based intervention aims to enhance use of guideline-based treatments [13–16]. In addition, these data are important because very little is known about patterns and variability in OA treatment in the primary care setting. Although some studies have described OA-related care in other countries [35–37], data are sparse for the U.S., particularly with respect to non-pharmacological care [38–40].

A novel and important component of the PRIMO study is the inclusion of a large number of primary care clinics. To our knowledge this is the first US-based clinical trial to involve a provider-based intervention for OA across multiple primary care sites. The PRIMO study clinics varied in important aspects including practice type (Family vs. Internal Medicine), size, and geographic setting. It is particularly important that the PRIMO study included clinics that varied in terms of urban/rural location and poverty level, as these factors can have an impact on the availability of some components of OA treatment, such as physical therapy services. Indeed, we found that some components of self-reported OA treatment use, particularly physical therapy for knee OA, tended to be lower among the three PRIMO clinics in rural locations (e.g. clinics #1, #2 and #5). This highlights the need for strategies to provide physical therapy services for patients in these underserved areas, since this is a core treatment component for knee OA.

Our data on recruitment processes illustrates the considerable effort involved in enrolling patients for this pragmatic clinical trial. Many clinical trials of behavioral interventions for patients with OA have utilized primarily self-referral mechanisms for enrollment [52–56]. We chose to actively enroll patients, via recruitment letters and telephone calls, because self-referral mechanisms may attract the most motivated patients, which may affect generalizability and overestimate intervention effectiveness when applied in real-world clinical settings. However, the type of active enrollment we utilized was very time and labor intensive. As shown in Table 2, 17% of patients who were invited to participate in the study via letter were randomized into the study.

Our recruitment process also differed from most prior OA clinical trials with respect to verification of diagnoses. Prior studies have typically either conducted de novo imaging to verify an OA diagnosis or relied solely on self-report of OA. The former is often impractical for large health services trials like PRIMO, and because we were delivering OA-specific recommendations to primary care providers, we chose not to rely exclusively on self-reported OA diagnosis, which may not have a high degree of accuracy. Therefore, study team members spent a considerable amount of time reviewing medical records to ensure that patients had a prior OA diagnosis, based on prior imaging (or, for knee OA, physician diagnosis plus meeting clinical criteria at the baseline visit [41]). As shown in Table 2, the study team reviewed over 16,000 medical records, with only 19% meeting initial eligibility criteria. One factor contributing to the high volume of records reviewed was inconsistent coding of OA in the medical record, resulting in the need to review records with a broad range of codes (e.g., “joint pain” instead of “osteoarthritis”) that encompass many other conditions. Some prior studies have used algorithms to identify patients with OA and other rheumatic conditions based on medical record data. Among algorithms used to identify people with OA, the specificity has been lower than would be optimal to rely solely on this information for inclusion on a clinical trial. However, these algorithms could be used as an initial step to identify patients who likely have OA, with additional screening conducted only for those individuals; this may be a more efficient approach that reduces the time required for medical record review [57]. More reliable coding of OA in health care settings would also have a tremendous positive impact on researchers’ ability to efficiently identify patients with this health condition for large health services studies.Our recruitment data also illustrate considerable heterogeneity across study clinics. First, there were differences in the proportions of patients who were eligible based on medical record review, ranging from 10%-31%; this was due to differences both in the presence/absence of OA diagnoses and exclusionary health conditions. Second, although the overall randomization rates of those mailed recruitment letters were relatively similar across clinics (13%-21%), there was variability in terms of whether more patients were found to be ineligible (14%-26%) vs. refused participation (48%-68%). As noted above, in order to meet recruitment goals at some clinics, we expanded the list of ICD-9 codes and/or the time window for time of last visit at the clinic. Based on conversations with providers at these clinics, we believe the need to expand the list of ICD-9 codes was largely a reflection of variations in coding practice; some providers tended to use more general joint pain related codes (e.g., 719.xx or 729.5) vs. OA-specific codes (715.xx) regardless of the severity or stage of disease. Although it is possible that use of different code sets across clinics could have led to variation in OA severity we did not see clear evidence of this with respect to baseline WOMAC scores. Specifically, the three clinics at which we expanded to more general codes were not among those with the lowest symptom severity (Figure 3). There were only two clinics in which we did not need to expand the time window since the prior clinic visit from 12 to 18 months. Patients in these two clinics overall may have had more frequent visits to providers, possibly indicating overall poorer health. In one of these clinics (#5), we did find the highest proportion of study participants with “fair or poor” general health. Given the overall heterogeneity in patient health status across clinics, it will be important for future analyses of this study to consider these variations in statistical models.

There was also considerable variability in patient demographic and clinical characteristics across PRIMO clinics, further enhancing generalizability of the study. With respect to demographic characteristics, there was substantial between-clinic variation in age, gender, race, and income status. It is particularly important that this study included clinics with high proportions of non-white patients and those with low income; these two demographic groups have been under-represented in OA clinical trials, and some studies have shown that these groups have lower referrals to and use of some OA-related treatment components [17, 38, 39, 58–60]. Clinic samples also varied in terms of self-reported overall health, with 20%-35% rating their health as fair or poor. This is important because overall health status may impact response to the patient behavioral intervention or the effectiveness of the provider-based intervention. In particular, it may be more challenging for providers to adequately address OA-specific issues for patients who also have multiple comorbid conditions. With regard to OA-specific characteristics, there was not much between-clinic variation in presence of knee OA (≥90% at all clinics), but there was substantial between-clinic variation in diagnosis of hip OA (26%-68%) and diagnosis of both knee and hip OA 23%-61%. Some clinics with a higher prevalence of hip OA were also those in which the average participant age was higher (e.g. clinics #2, #3, and #7). Not surprisingly, the duration of arthritis symptoms was also longer in several of the clinics with the highest average patient age (e.g. clinics 2, 3, and 4). There was between-clinic variation in WOMAC scores, ranging from 34.7-44.2 , on a scale of 0–96; this range likely reflects clinically meaningful differences in average symptom severity across these clinic samples [61].

Finally, we also observed substantial between-clinic variation in some of the self-reported OA treatment use variables. Although the majority of patients at all clinics reported using pain medications for managing OA, there was notable between-clinic variation for use of some specific medication classes. This is not surprising since there is no standard algorithm for pharmacological treatment of OA [13, 15, 62], and patients have differential response to various pain medications. There were also variations in use of non-pharmacological therapies. In particular, there was substantial between-clinic variation in patients who reported ever seeing a physical therapist, both for knee and hip OA. The evidence base for physical therapy as a treatment for knee OA is well established [13–16], but less than 1/3 of patients at half of PRIMO study clinics reported ever having received this treatment, despite the fact that average symptom durations were 8–13 years. In addition, although 40%-64% of patients with knee OA reported using some type of knee brace, much smaller proportions (0%-18%) reported using a more supportive type of brace with metal supports. Although there is some conflicting evidence regarding the effectiveness of braces for knee OA [63], these results may reflect underutilization of this treatment option. We did observe a tendency for patients at some clinics to report greater use across different categories of non-pharmacological treatments. For example, clinics #9 and #10 had among the highest proportions of patients reporting receipt of physical therapy and knee braces (both in general and with metal supports), as well as receipt of knee and hip injections. Interestingly, these two clinics also had the lowest proportions of patients reporting fair or poor health, suggesting that OA-specific treatments may be more highly utilized among patients with fewer competing health problems. Overall, PRIMO data reflect significant variation in the use of OA therapies – particularly non-pharmacological ones - across different primary care settings. This may be partly due to the lack of practical guidance in existing OA treatment recommendations for when particular treatments are best utilized [64].

There are several limitations to this study. First, all data were self-reported, and there may be errors in patients’ recall of past OA treatment use. However, there is not a reliable method for obtaining objective data of this nature, and we have no reason to suspect that recall bias would be differential across clinic sites. Second, although we believe the practices chosen for this study are representative of broader primary care practices in the study region, all practices were part of one health care organization, and it is possible that these clinics, their patient populations and their patterns of OA treatment may differ from other health care settings. Similarly, the recruitment-related data describe our experiences with this particular clinical trial, and other studies may differ in rates of patient eligibility, enrollment and other process variables. Third, we described the number and types of providers at the time enrollment began at each clinic, and in some cases there may have been subsequent personnel changes. However, we know of no large changes in numbers of providers at PRIMO clinics during the study period. Fourth, although we asked participants whether they had ever received physical therapy care for knee or hip OA, we did not assess whether participants had received exercise instructions from other providers. Therefore we may have underestimated the proportions of patients who received some exercise-related recommendations, though other research suggests this type of patient education may be infrequent in primary care [26]. Fifth, in this pragmatic, effectiveness study we did not obtain new radiographs or MRIs to ascertain OA diagnoses but rather recruited patients who had already received these OA diagnoses as part of routine clinical care. Therefore specific criteria used to diagnose OA based on imaging may have differed across clinicians. Sixth, because OA is not always documented in the medical record, some patients in these Duke healthcare system clinics with hip or knee OA may have been inadvertently excluded from potential contact for this study.

In summary, these data show both the challenges and important advantages of conducting practice-based research, particularly including diverse primary care clinics. Although PRIMO study clinics were part of one health care system, they varied considerably, as did participants enrolled in the study. It is particularly important that we were able to enroll participants that differed in terms of overall health status, income, OA symptom severity, and baseline OA treatment use, as these may all impact intervention response. Inclusion of differing clinics will also allow assessment of practical aspects of implementing a provider-based OA intervention across different primary care settings. Finally, these results illustrate specific challenges of conducting pragmatic trials that actively recruit patients based on electronic medical record data. Two main challenges were variability in diagnostic coding and the volume of time required to review medical records for eligibility criteria. Although this study focused on OA, these challenges would likely apply to many other chronic health problems since coding variability exists for many conditions [65–68]. The increased availability of electronic medical records holds great promise for clinical and health services research, but there is a need for more efficient and standardized processes that facilitate identification and recruitment of patients for these kinds of pragmatic studies.