Uptake of evidence by physicians: De-adoption of erythropoiesis-stimulating agents after TREAT trial showed they are ineffective and unsafe

Variation in de-adoption of ineffective or unsafe treatments is not well-understood. We examined de-adoption of erythropoiesis-stimulating agents (ESA) in anemia treatment among patients with chronic kidney diseases (CKD) following new clinical evidence of harm and ineffectiveness (the TREAT trial) and the FDA's revision of its safety warning. We used an interrupted time series approach to estimate changes in use of epoetin alfa (EPO) and darbepoetin alfa (DPO) in the commercial and Medicare Advantage (MA) and Medicare fee-for-service (FFS) populations. We also examined how changes in both trends and levels of use were associated with physicians’ characteristics. Study cohort included patients with CKD stages 3 to 5 during 2007-2015. Use of DPO and EPO declined over the study period. There were no consistent changes in DPO trend across insurance groups, but the level of DPO use decreased right after the FDA revision in all groups. The decline in EPO use trend was faster after the TREAT trial for all groups. Nephrologists were largely more responsive to evidence than primary care physicians. Differences by physician's gender, and age were not consistent across insurance populations and types of ESA.


Abstract
Variation in de-adoption of ineffective or unsafe treatments is not well-understood. We examined de-adoption of erythropoiesis-stimulating agents (ESA) in anemia treatment among patients with chronic kidney diseases (CKD) following new clinical evidence of harm and ineffectiveness (the TREAT trial) and the FDA's revision of its safety warning. We used an interrupted time series approach to estimate changes in use of epoetin alfa (EPO) and darbepoetin alfa (DPO) in the commercial and Medicare Advantage (MA) and Medicare fee-forservice (FFS) populations. We also examined how changes in both trends and levels of use were associated with physicians' characteristics. Study cohort included patients with CKD stages 3 to 5 during 2007-2015. Use of DPO and EPO declined over the study period. There were no consistent changes in DPO trend across insurance groups, but the level of DPO use decreased right after the FDA revision in all groups. The decline in EPO use trend was faster after the TREAT trial for all groups. Nephrologists were largely more responsive to evidence than primary care physicians. Differences by physician's gender, and age were not consistent across insurance populations and types of ESA.

Background
There is a growing interest in understanding physicians' decisions to discontinue use of ("de-adopt") treatments in light of new clinical evidence suggesting that a previously-approved treatment is ineffective or unsafe. [1][2][3] While existing studies mostly focus on measuring the rate of de-adoption or reductions in prescribing, there is less evidence regarding variation in de-adoption. 3,4 Of particular interest are the impact of physician attributes and patient insurance type. Previous research on physicians' adoption of new treatments based on clinical evidence suggests that physicians' decisions are in uenced by various factors such as physicians' specialization and experience as well as patients' health insurance. [5][6][7][8][9][10][11] However, the de-adoption process is not simply a mirror re ection of the adoption process and, hence, may not be in uenced by the same factors or in similar manners. 2 In this study, we examined the de-adoption of erythropoiesis-stimulating agents (ESA) in the treatment of anemia among patients with chronic kidney disease (CKD) and subsequent changes in prescribing. Anemia is common among CKD patients, and ESA treatment is used to stimulate bone marrow to produce red blood cells, preventing the need for blood transfusion. 12 ESA treatment is typically triggered when a patient's level of hemoglobin concentration (measured in gram per deciliter, or g/dL) is too low, and the treatment is used until the hemoglobin concentration reaches a "safe" range for patients. 13 The target range has changed over time as recent clinical studies have highlighted the potential danger of setting the target hemoglobin level too high. 14,15 The main types of ESA in the U.S. are epoetin alfa (EPO) 16 and darbepoetin alfa (DPO), 17 which mainly differ in how frequently the drug is administered to patients. 13 DPO is a newer, synthetic form of naturally-occurring erythropoietin that has a longer duration of action, requiring less frequent administration; 18 EPO is usually administered three times a week while DPO is administered once a week up to once every a month. 19,20 The rst study to show evidence of unsafety and ineffectiveness of ESA use was the Trial to Reduce Cardiovascular Events with Aranesp Therapy (TREAT) published in October 2009. The trial found that targeting a hemoglobin level of 13 g/dL using DPO did not reduce mortality, cardiovascular, or renal events, but resulted in a higher risk of stroke, compared to using DPO when the hemoglobin level fell below 9 g/dL. Following this publication, the Food and Drug Administration (FDA) revised its original black-box warnings in June 2011, recommending to only start ESA treatment when the hemoglobin level falls below 10 g/dL, with the lowest dose necessary to avoid blood transfusion.
Several studies have used administrative claims data to examine changes in overall ESA prescribing following the TREAT trial publication and the FDA revision of the original warnings. Thamer et al. 21 compared ESA prescribing among Medicare patients with CKD stages 3 and 4 before and after the TREAT trial publication and found that prescribing rates dropped signi cantly; trend analysis suggested that ESA prescribing had been declining before the publication, but the decline accelerated following the event. Similarly, Park et al. 22 found that ESA prescribing among patients in the Truven MarketScan database with CKD stages 3 to 5 dropped signi cantly following the trial and the FDA revision. Finally, Sarpatwari et al. 23 evaluated changes in initiation of ESA among patients with cancer after a Risk Evaluation and Mitigaton (REMS) program was implemented and did not nd signi cant changes in trends. However, neither study examined physician patterns of deadoption, nor whether these patterns differed by insurance type. If certain types of physicians or certain pharmacy plans were associated with lower rates of de-adoption, the information could guide efforts to reduce the use unproven treatments. Moreover, neither study separated the impact of the evidence on the use of EPO and DPO. Therefore, we examined ESA de-adoption among advanced CKD stages 3-5 patients -those who are more likely to have anemia relative to patients with early stages of the disease. 24 First, we examined de-adoption of EPO separately from that of DPO; while the TREAT trial was about DPO treatment, the FDA revision applied to both EPO and DPO, and it is plausible that the impact of the TREAT evidence differed between the two. Second, we examined which physician characteristics were associated with de-adoption, examining changes in prescribing in both levels and trends in response to both the trial and the FDA revision. And third, we examined de-adoption in multiple insurance populations, namely commercially insured, Medicare Advantage (MA), and Medicare Fee-for-Service (FFS) bene ciaries, and assessed whether physician attributes had different associations across these payers.

Data source
We used administrative claims data from two sources. The rst source was the 2007-2015 commercial and MA administrative claims from the OptumLabs® Data Warehouse (OLDW), a comprehensive, longitudinal, real-world data asset with de-identi ed claims and clinical information. 25 The second source was a 20% random sample of Medicare Fee for Service (FFS) bene ciaries from 2007-2013. We combined the administrative data with information about physicians from Doximity®. Doximity is a data resource that allowed us to observe key physician characteristics. This database includes information from a wide range of sources, such as the National Provider Identi er Registry and state medical boards, and has been validated and used in previous literature. 26 − 29 Study population We identi ed three separate cohorts of patients who had at least one claim for CKD diagnosis (based on ICD 9-CM and ICD 10-CM diagnosis codes) anytime between 1/2007 and 12/2015; all were identi ed using medical claims in either commercial, MA or FFS data sources. We restricted the cohorts to those who had continuous medical and prescription drug coverage for the 12 months before the index diagnosis claim for CKD. In any given month, we agged a patient to have CKD if they had at least one inpatient claim or two outpatient claims spaced more than 30 days apart with CKD diagnosis in the past 6 months. Our nal analytical samples consisted of the patient-months identi ed to have CKD stages 3-5 without dialysis treatment. Supplemental Digital Content, Sect. 1 provides details on the ICD 9-CM and ICD 10-CM diagnosis codes and CPT codes used to identify the study sample.

Physician Assignment
To analyze changes in ESA use by physician characteristics, we attributed each patient-month observation to the physician responsible for making decisions concerning the ESA treatment following a 2-step procedure. In the rst step, we isolated medical claims for all evaluation and management (E&M) visits for each patient-month observation and divided claims into four categories based on the specialty of the associated physician: (1) nephrology, (2) internist, (3) hematology and oncology, and (4) all others. Drawing from the literature, 30 we ranked these categories based on relevance to anemia treatment for patients with CKD with (1) being the most relevant and (4) being the least relevant specialty. For example, even if an internist has more E&M claims than a nephrologist in a given month, the nephrologist is attributed to the patient-month. Within each specialty category, the physician with the most E&M visits was attributed to the patient-month (ties were assigned randomly); and we carried forward that attribution, month after month, until an interruption in patient-months at risk or a change in physician with the most visits. Once each patient-month observation was attributed to a physician, we used the National Physician Identi cation (NPI) to merge in physician characteristics from the Doximity → database.

Measures
The two outcome variables were dichotomous indicators for DPO and EPO use in a given patient-month observation (unit of analysis). The indicators were constructed by identifying ESA administration in outpatient claims (CPT codes "J0881" (Darbepoetin Alfa) and "J0885" (Epoetin Alfa)) or in pharmacy claims. Patient covariates included sex, age, CKD stage, and Elixhauser comorbidity index. Physician characteristics included sex, specialty, and age (< 50, 50 or older), and physician specialty (primary care physicians (PCPs), nephrologists, and non-nephrology specialists, including internists, hematologists, and oncologists).

Statistical Analysis
We summarized DPO and EPO use, patient and physician characteristics, reporting mean (SD) or n (%) according the characteristic. Then, to assess patterns of de-adoption we estimated a series of mixed effects models. First, we examined changes in ESA use among CKD patients after new evidence of unsafety and ineffectiveness from the TREAT trial publication (10/2009) and the FDA revision (6/2011), using standard interrupted time series (ITS) models. [31][32][33] We considered three time-periods based on these events: We excluded the three-month periods before and after each event to avoid capturing any anticipatory or short term effects. 33 We estimated changes in the level and the trend in use of DPO and EPO use in period 1 relative to the baseline, as well as in period 2 relative to period 1. All models adjusted for patient and physician characteristics and included calendar month dummy variables to account for seasonality. Standard errors were clustered at the patient level. Models were estimated separately for commercially insured, MA, and Medicare FFS cohorts. These models let us assess whether there were differences in de-adoption by payer; though data use agreements precluded combining these cohorts for formal testing, we report the effect magnitudes and P-values for each. Then, to assess whether de-adoption (changes in levels or trends) differed by physician characteristics, we estimated a second set of models which included interactions of the levels and trends with physician characteristics one at a time (for example nephrologist or not). Details of the model speci cations are provided in Supplemental Digital Content, Sect. 2.
All analyses were performed with SAS, Version 9.4 (Copyright © 2002-2012 SAS Institute Inc.) and Stata 14.2 (StataCorp, College Station, TX). The study was deemed exempt from review by the University of Minnesota Institutional Review Board because the data were deidenti ed.

Results
Our study included 501,287 patient-month observations for the commercially insured, 1,206,050 for MA, and 17,405,319 for Medicare FFS. Unadjusted rates of DPO use were 5.3%, 3.2% and 2.6% for each insurance group respectively, while corresponding unadjusted rates of EPO use were 7.2%, 5.3% and 3.1% (Table 1). For the commercially insured, the mean (SD) age was 61.7 (13.1) and the mean Elixhauser comorbidity index was 6.  For all three cohorts, unadjusted rates of EPO and DPO use were declining before the TREAT trial and over the entire study period (Fig. 1  Before the TREAT trial publication, EPO use was decreasing for commercial and Medicare FFS patients (both, P < 0.001), but was at for MA patients. After the publication, there was no change in levels for commercially insured and MA but an increase in level for Medicare FFS (P < 0.001). EPO trends decreased further for all three cohorts (all P-values < 0.001). Following the FDA revision, MA and FFS patients experienced a decrease in levels (both P-values < 0.05), and all patient groups experienced an increase in trends (all P < 0.005).
Changes in trends and levels of use over time by physician characteristics DPO (Table 3)  Note: Estimates are reported for changes of trends and levels of DPO use (compared to previous periods) by physician's characteristics as well as P-value for differences across physician characteristics. Baseline refers to the period between Jan-2007 to June-2009; period 1 (post-TREAT/pre-FDA) is between Feb-2010 and Feb-2011; period 2 (post-TREAT/post-FDA) is between Oct-2011 and Dec-2015. All estimates were multiplied with 100 to represent percentage point changes. All samples consist of patients with CKD stage 3 to 5. All models control for patients' sex and age and physicians' specialty, age, sex, and experience (see text for details). Standard errors are clustered at the patient level; 95% con dence interval is reported in parentheses, and P-value is reported in brackets.
In Medicare FFS, levels in DPO use increased immediately following the trial publication, and decreased following the FDA revision (all Pvalues < 0.05) both for nephrologists and PCPs. There were no immediate level changes in commercially insured and MA for either the nephrologists or PCPs.
Monthly DPO trends increased after TREAT publication for nephrologists in all three insurance cohorts, but increased for PCPs only in the Medicare FFS cohort (all P-values < 0.05); however there was no signi cant difference in trends between these two specialty groups in any cohort. After FDA revision, monthly trends increased more for nephrologists in the FFS group, but all other trends were stable.
Levels of DPO use increased immediately following the trial publication similarly for male and female physicians, but only in Medicare FFS (P-values < 0.01). Following the FDA revision, levels of use decreased for male physicians across all insurance groups (P-values < 0.01); and decreased for female physicians only in Medicare FFS (P-value < 0.001).
Trends in DPO among male physicians increased (became less negative) in commercial and Medicare FFS groups after the trial publication (P-value < 0.01) and remained stable in MA group. Among female physicians, trend increased only in Medicare FFS (P-value < 0.001). Changes in trends after the FDA revision was also largely statistically insigni cant except for increases in MA among female physicians, and in Medicare FFS among male physicians (P-values < 0.01).
There was a decrease in DPO levels following trial publication and the FDA revision for physicians aged 50 and older, but only in commercially insured (P-values < 0.05). In Medicare FFS, levels increased similarly in both age groups following the trial publication, but decreased following the FDA revision (all P-values < 0.001).
There were no changes in DPO trends by physician age across the three insurance groups. Only in Medicare FFS, we observed that physicians aged 50 and older had a larger increase in DPO trend after the trial publication (P-value of difference = 0.037). However, this relationship was reversed after the FDA revision with the older physicians having a smaller increase in trend compared with the younger physicians.
EPO (Table 4) Note: Estimates are reported for changes of trends and levels of EPO use (compared to previous periods) by physician's characteristics as well as P-value for differences across physician characteristics. Baseline refers to the period between Jan-2007 to June-2009; period 1 (post-TREAT/pre-FDA) is between Feb-2010 and Feb-2011; period 2 (post-TREAT/post-FDA) is between Oct-2011 and Dec-2015. All estimates were multiplied with 100 to represent percentage point changes. All samples consist of patients with CKD stage 3 to 5. All models control for patients' sex and age and physicians' specialty, age, sex, and experience (see text for details). Standard errors are clustered at the patient level; 95% con dence interval is reported in parentheses, and P-value is reported in brackets.
There were no EPO level changes after the TREAT trial for either the nephrologists or PCPs; the level decreased after the FDA revision for nephrologists, but only in Medicare FFS (P-value = 0.03). After the TREAT trial, EPO monthly use trend decreased for nephrologists in all insurance groups (P-values < 0.01). In contrast, for PCPs, decrease in trend was observed only in Medicare FFS (P-value < 0.01). Following the FDA revision, trends increased among nephrologists in all three insurance groups (P-values < 0.001). Among PCPs, increase was observed only in Medicare FFS (P < 0.001).
Among female physicians, there were no changes in EPO levels across insurance groups after the TREAT trial or the FDA revision. However, among male physicians, levels decreased in MA and increased in Medicare FFS after the trial publication (P-values < 0.05), and decreased in both insurance groups after the FDA revision (P-values < 0.05).
The change in trends after TREAT trial was signi cantly different between male and female physicians only in MA. While the trend declined for male physicians (P-value < 0.001), it remained stable for female physicians (P-value = 0.741), a relative difference signi cant at P-value = 0.017. There were no signi cant differences in the change of trends following the FDA revision.
Among physicians aged 50 and older, there were no changes in EPO levels across the board except in Medicare FFS; an increase after the TREAT trial (P-value = 0.007) and a decrease following the FDA revision (P-value = 0.016). Among the younger group of physicians, levels declined in MA and increased in Medicare FFS after the TREAT trial (P-values < 0.01), and declined in MA following the FDA revision (P-value = 0.005).
Older and younger physicians had similar changes in trends after the TREAT trial, and after FDA revision with no signi cant differences in relative change in trends except for Medicare FFS where physicians aged 50 and older had slightly lower increase in trend after the FDA revision relative to younger physicians (P-values < 0.001, a difference signi cant at P = 0.025).

Discussion
In this examination of ESA use in three insurance cohorts, we found that DPO and EPO use were both already declining prior to the TREAT trial publication, and they continued to decrease following the TREAT trial and the FDA black box warning revision. While this was consistent with prior research, we also found differences in how the trends and levels changed: by treatment, by insurance group and by physician characteristics. Consistent with expectations, the decline in EPO use became steeper after the TREAT trial across all three insurance groups, but surprisingly, the decline in DPO use slowed (the trend was less negative) after the TREAT trial in commercially insured and Medicare FFS groups, and was unaffected in MA group. Moreover, after the FDA revision, the trends in DPO use increased again in the MA and Medicare FFS, though not in the commercial group, while the trend in EPO use increased in all three groups.
Notably, although the DPO trend increased after both the TREAT trial and the FDA revision, DPO use overall dropped after the FDA revision in all three insurance groups; this suggests both that the subsequent weaker decline (relative to the trend prior to the FDA revision) may re ect in part lower overall use, and also that the FDA revision is viewed as a stronger evidence relative to the TREAT trial publication associated with a decrease in use. We observed the similar immediate decline for EPO use after the FDA revision as well.
Differences in de-adoption across insurance cohorts were minor and with no consistent relationship between insurance group and changes in levels or trends. That we found fewer changes in levels and trends in the commercial cohort may re ect the smaller sample size, especially compared with the FFS group (with 34 x as many patients), but even that cohort had more than half a million patients. More likely these differences re ect differences in providers who treat more or fewer commercial, FFS or MA patients.
With regards to providers, we found that both the less negative trend in DPO use as well as the steeper decline in EPO use after the TREAT trial were generally driven by nephrologists to a greater extent. We also observed some differences in responses to the trial and the warning revision by physician's gender and age, but the differences were not consistent for DPO and EPO and for different insurance populations.
These results suggest that physician specialty have a dominant role in prescribing decision, and that specializations with higher use of treatment were more responsive to new evidence of unsafety and ineffectiveness. This is likely because the patient populations attributed to specialists for a given condition, in this case nephrologists, may be more severe in unobserved ways.
Our ndings are broadly consistent with prior research on de-adoption of treatments, which has found variation across providers in how rapidly treatments fall out of use in the wake of new evidence. Borne et al. 34 found that variation across institutions increased with increasing de-adoption of de brillation following evidence of risk, indicating that providers de-adopted at different rates; they attribute this to different institutional practices. Bekelis et al. 4 found that more experienced physicians reduced their use of carotid revascularization more quickly than other physicians, while higher volume physicians reduced their use more slowly. More generally, van Dulmen et al. 35 identi ed 263 barriers to de-adoption of treatments, of which the majority were physician factors. Thus, there are large variety of factors that may systematically in uence the de-adoption of EPO and DPO, which suggests the need for systematic efforts to promote de-adoption. 36 There are several limitations this study. Our patient populations were identi ed using diagnosis codes and medication use in administrative claims. Moreover, our analyses could not account for many unobserved factors of ESA use such as treatment cost, patient preference, and ESA treatment appropriateness. The FDA warning indicated that ESA treatment can be considered if the hemoglobin level was less than 10 g/dL, the rate of hemoglobin declined, and reducing the risk of alloimmunization and/or red blood cell transfusion was the goal. These unobserved conditions would in uence physician decisions whether to prescribe ESA treatment for patients.
In summary, we found that the use of DPO and EPO started decreasing before new evidence on their ineffectiveness and unsafety, and continued to decrease over the study period. However, the impact of the TREAT trial and of the FDA revision was limited and inconsistent, with a trend that decelerated for DPO for some payers. Our ndings indicate that de-adoption of ineffective and unsafe treatments varies across insurance type and across physician specialty and characteristics. It is important to take these variations into account when designing practice guidance to promote e cient de-adoption.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.