The Association of Provider and Practice Factors with HIV ART Adherence

Abstract

Objective

While Anti-Retroviral Therapy (ART) is essential to patients with HIV, there is substantial variation in adherence nationally. We assess how provider and practice factors contribute to successful HIV ART adherence.

Design

We used Medicaid Analytic Extract claims from 2008–2012. We attributed patients with HIV to the provider that provided the plurality of HIV-related services or primary care in a given year and assigned these providers to a medical practice based on the National Provider Identifier registry file. We fit successive linear hierarchical models with patient, provider, and practice characteristics to partition the variation in adherence driven by each factor. Our unit of analysis was the patient-year.

Setting

14 US states with the highest HIV prevalence

Participants

111,013 patient-years representing 60,496 Medicaid enrollees living with HIV attributed to 4,930 providers and 1,960 practices

Main outcome measure

Percent of year individual patients were adherent to an ART regimen

Results

Provider and practice random effects jointly explained 6.8% of variation in adherence with patient differences accounted for 45.2% of the variation. Patients seen by generalists and other specialists had a 1.6 (95%CI: 0.6–2.5) and 5.1 (95%CI: 4.1–6.1) percentage point greater adherence than those seen by infectious disease specialists (p<0.001). Every additional year a patient saw the same provider was associated with a 6% increase in adherence (95%CI:5.7–6.3).

Conclusions

There is substantial variation in ART adherence attributable to providers and practices and between provider specialties. To improve ART adherence for patients living with HIV, structural aspects of care should be considered.

INTRODUCTION

Timely provision of antiretroviral therapy (ART) to persons living with HIV (PLWH), and the long-term maintenance of such therapy, remains the cornerstone of high quality HIV care [1,2]. Beginning with the advent of modern ART in the mid-1990s, there has been abundant research focusing on predictors of non-adherence [3–5], but the vast majority of research has focused on patient-level factors such as age, gender, race, depression, and substance use. While person-level factors are critical, less attention has been given to characteristics of the health care system within which PLWH get their care, such as the role of providers or provider practices. These system-level, structural factors may be important drivers of patients’ ART adherence, and may be more amenable to interventions than many patient factors.

Some previous work has focused on the role of provider training and expertise on outcomes in PLWH. HIV expertise has been shown to be linked to HIV caseload [6], and generalist and infectious disease specialists with high HIV volume are equally likely to get their patients on ART [7]. In a study of HIV care quality, well-trained nurse practitioners provided care quality that was equal to that of physicians [8]. One study found wide differences in site-level ART prescribing, and that higher site HIV volume was associated with increased ART prescribing [9]. Others have documented large heterogeneity in rates of ART adherence across sites in the US [10] and sub-Saharan Africa [11]. However, most of these studies were done before modern ART, few examined physician and site characteristics simultaneously, and only one used multilevel modelling techniques [11].

We used a national sample of Medicaid patients from 2008–2012 to examine provider and care practice characteristics associated with ART adherence. The Medicaid population living with HIV is particularly important because Medicaid is the largest source of insurance among PLWH, covering over 40% of all PLWH who receive regular care [12]. We had two main study questions. First, what fraction of variance in patient-level ART adherence can be explained by provider and practice-level effects after adjustment for patient-level factors? Second, in multilevel models, what provider and practice characteristics are associated with ART adherence?

METHODS

Data Sources

Our primary data source was Medicaid Analytic Extract (MAX) claims from 2008–2012. MAX data provide 100% of Medicaid claims from each state, and are the only currently available source of Medicaid data that is unified across states. While our dataset ends in 2012, this represents the most recent year of complete data from these states as of the time of this study. For our analysis, we had data from 14 states (California, Florida, Georgia, Illinois, Louisiana, Massachusetts, Maryland, North Carolina, New Jersey, New York, Ohio, Pennsylvania, Texas, Virginia). These states were chosen because they are the 14 states with the highest HIV prevalence, accounting for approximately 75% of all HIV cases in the US. We used the MAX Prescription Drug file to determine use of ART therapy, the Other Therapies (OT) file to determine use of outpatient services, primary care, and specialist office visits, and Inpatient Therapies (along with OT files) to determine a comorbidity count. We used the National Provider Identifier (NPI) registry file[13] published by CMS in order to attribute patients to their primary HIV care providers, and for provider characteristics.

Study Sample

Our sample included patients aged 18–64 with HIV in the 14 states from 2008–2012. We classified a patient as having HIV if they had at least 2 claims at least 30 days apart that listed HIV as a diagnosis, or if they used ART, as ARTs are seldom prescribed for patients without HIV. This classification method has been used in prior work [14,15]. We excluded patients who were dually enrolled with Medicare, as we did not have Medicare claims available for this study, and those that were enrolled in Medicaid managed care, as the MAX claims may not contain complete managed care claims during the time period of interest. We required at least 6 months of Medicaid enrollment in a year to be included in our sample. Note that because patients were included based on ART use, only patients judged by their clinicians to be eligible for ART were studied.

Provider and Practice Attribution

For each calendar year of the study, we assigned patients to a primary HIV care provider (PCP-HIV). To accomplish this, we designed a multi-stage attribution process based on similar approaches used in the literature [16]. We assigned patients to the providers that provided the plurality of their ART prescriptions and primary care visits. We then attributed providers to practices on the basis of information from the NPI registry file. A detailed description of the attribution process can be found in Appendices 1 and 2. In this study, we define providers as either physicians, Physicians Assistants, or Nurse Practitioners.

Outcome

Our primary outcome of interest was the percent of the calendar year that PLWH were exposed to ART. To calculate this, we use days supplied and fill dates from the MAX prescription file to count the number of days in the year the patient had ART available. This number became the numerator. The denominator was 365 days if the patient was alive and enrolled for the whole year (e.g., if the patient was enrolled in Medicaid for half the year, we would only count 6 months towards the denominator). This is often referred to as Proportion of Days Covered (PDC). More details on the calculation of the measure are found elsewhere [17].

Patient and provider characteristics

Patient characteristics included age, gender, race/ethnicity, and basis of Medicaid eligibility. Patients were classified as having a chronic condition if they had two outpatient codes, or one inpatient code, for a diagnosis. We also classified patients as having a problem with drug use, alcohol use or both. Importantly, substance use claims were not redacted from our data. From the prescription file, we included a flag for what ART regimen the patient was primarily receiving during the year (NRTI, NNRTI, Boosted PI, Integrase, PI, multiple/others) as regimen may be related to adherence. Based on the attribution, we further included the consecutive number of years the patient was attributed to the same provider.

At the individual-provider level, we included variables from the NPI registry file for primary taxonomy/specialty (Generalist vs. Infectious Disease vs. Other Specialty) and credential (MD/DO vs. Nurse practitioner/physician assistant/other). A full list of specialty classifications are included in Appendix 6. We also included number of patients that were attributed to a given individual-provider as a measure of experience treating patients with HIV, recognizing that the count of Medicaid patients only partially captures a provider’s HIV patient load. At the practice-level, we examined the number of patients and number of providers attributed to that practice.

Statistical Methods

Our primary unit of analysis was person-year. We examined the distribution of patients’ adherence at the provider and practice-level. Based on the mean adherence of patients attributed to a provider or practice, we plotted the distribution of adherence across providers or practices. We then fit a linear model adjusting for age, gender, race/ethnicity, regimen, chronic conditions, substance use, and state, year, and provider or practice fixed effects. From this model, we calculate the adjusted adherence level for each provider or practice and also plotted this distribution.

In our primary analysis, we fit multiple hierarchical models on the patient-year level data with PDC as the outcome, and with provider and practice identifiers specified as random effects in nested multi-level models. The use of random effects in multi-level models allowed us to account for the natural clustering of patients within providers and practices, and enabled us to partition the variance in patient outcomes that can be attributed to the patient versus the provider [17–23]. We fit two unconditional models, the first with provider random effects alone, and the second with practice random effects alone. Next, we included both provider random effects nested within practice random effects. We then added provider and practice characteristics detailed above, and in our final model also included patient characteristics and a patient random effect to account for multiple observations from the same patient over time. From each subsequent model, we calculated the variance partition coefficient, which is the percent of the total variance in the outcome explained by the provider or practice, accounting for all other included patient characteristics. We assessed how much variance was due to the provider practice, or patient, and using the full model, measured the variables that were most associated with adherence outcomes. All models were linear models with state and year fixed effects to adjust for differences in state Medicaid programs, and used robust standard errors. As a sensitivity check, we log transformed the outcome and found similar results. To determine if our results were not purely driven by differences in geography, we also fit a model with zip code fixed effects to control for geographic access to different types of providers. An alpha of p<0.05 was considered statistically significant. All analyses were conducted in Stata Version 15 (StataCorp, Texas).

RESULTS

Sample characteristics

Over 80% of all HIV patients during the study period were successfully attributed to providers, with some variation between states in the rates of attribution (See appendices 3 and 4). Our final sample included 60,496 patients with HIV contributing a total of 111,013 patient years during our study period.

Table 1 displays patient characteristics for the overall sample and compares patients in the highest versus lowest quintiles of adherence performance at the provider-level. The percent of year adherent across all patients in the sample was 69.8% (SD 28.5%). Given our large sample sizes, almost all differences between patients in the lowest and highest quintiles were statistically significant, but many were also clinically large. Notably, patients attributed to low adherence providers were more often black (56.7% vs 35.9%), younger (46.4 vs 52.7 years old), disabled (50% vs 42.8%), and had fewer years on average attributed to the same provider (1.3 vs 2.4).

Table 1:

Characteristics of patients and providers by ART adherence performance

Variable	All Patients	Lowest PCP Adherence Quintile	Highest PCP Adherence Quintile	Significance of Difference
Patient Characteristics
n patients	60,496	3,414	10,207
n patient years	111,013.0	5,989.0	17,873.0
% Female	45.3	56.9	41.4	<0.001
% HMO	11.8	14.1	8.5	<0.001
% White	13.0	14.0	10.3	<0.001
% Black	44.0	56.7	35.9	<0.001
% Hispanic	19.6	10.2	33.2	<0.001
Median Age	50.2	46.4	52.7	<0.001
% Eligible on basis of disability	46.8	50.0	42.8	<0.001
Mean number of chronic conditions	6.9	6.9	7.5	<0.001
% with Alcohol Abuse	30.5	23.6	38.5	<0.001
% with Drug Abuse	40.9	34.4	51.4	<0.001
% with any Substance Abuse	43.9	39.1	54.2	<0.001
Mean Annual Adherence:	69.9	42.6	83.1	<0.001
Mean Years Assigned to Same Provider	2.1	1.3	2.4	<0.001
Mean Years Assigned to Same Practice	2.2	1.4	2.5	<0.001
Regimen
NRTI	9.2	11.4	5.0	<0.001
NNRTI	28.5	27.2	29.4
Boosted PI	40.5	41.9	41.9
Integrase	4.8	4.0	4.8
PI	4.5	5.1	5.7
Multiple/Others	12.5	10.4	13.2
% of Year Adherent (SD)	69.8 (28.5)	38.5 (10.6)	88.1 (6.1)	<0.001
Practice/Provider Characteristics
N of Providers	4930	986	986
Mean N of Patients to Provider (median)	8.2 (3)	3.4 (2)	7.3 (2)	<0.001
% Providers Female	33.0	31.3	30.3	0.63
% Specialty ID	19.3	18.2	12.3	0.001
% Specialty Generalist	32.8	32.1	31.0
% Specialist Other	47.9	49.8	56.7
% MD/DO	83.4	85.0	83.8	0.46
N of Practices	1960	392	392
Mean N of Providers to Practice (median)	3.6 (2)	2.9 (2)	2.9 (2)
Mean N of Patients to Practice (median)	18.0 (45)	6.0 (3.0)	18 .4 (3)	<0.001

Provider and practice level adherence performance

Figure 1 shows the distribution of unadjusted and adjusted adherence for each provider or practice. In the unadjusted model, the median adherence for providers was 66.2% with an interquartile range of 23.0 percentage points. For practices the median adherence was 66.2% with an interquartile range of 19.8 percentage points.

Figure 1: Distribution of Crude and Adjusted ART Adherence by Provider and Provider.

Notes: Un-adjusted is a boxplot of the distribution of provider or practice level ART PDC rates. It is comprised of the mean adherence across all patients attributed to each provider or practice. The adjusted distribution comes from a model at the patient level their adherence adjusting for age, gender, race/ethnicity, regimen, chronic conditions, substance use, and state, year, and provider or practice fixed effects, then calculating the adjusted mean for each provider or practice.

After adjusting for patient characteristics, for providers, the median was 68.8% with an interquartile range of 10.4 percentage points, representing a 16.2% difference between the 25^th and the 75^th percentiles. For practices, the median adherence was 68.7% with an interquartile range of 9.5 percentage points representing a 15.0% difference between the 25^th to the 75^th percentiles.

Results of multi-level modeling of patient-level adherence

Table 2 presents the coefficients from the full regression model that includes both provider and practice random effects. Each coefficient represents a percentage point difference in the percent of year adherent at the patient-level. At the provider and practice-level, patients seen by generalists and other specialists had a 1.6 (95% CI: 0.6–2.5) and 5.1 (95% CI: 4.1–6.1) percentage point greater adherence compared with those seen by infectious disease specialists (p<0.001). A provider having an MD/DO as their credential was also associated with a 1.6 percentage point increase in adherence (p=0.016). Patients who saw providers with greater HIV patient volume tended to have higher adherence; however, the size of the association was small: every increase in 10 patients was associated with a 0.4 percentage point increase in adherence.

Table 2:

Regression coefficients for Percent of Year Adherent in Mixed Effect Model

Outcome: Percent of Year Adherence	Coef.	Std. Err.	P>z	[95% Conf.	Interval]
Provider Characteristics
Provider Female	0.7%	0.0040	0.079	−0.1%	1.5%
Provider has MD/DO	1.6%	0.0065	0.016	0.3%	2.9%
+10 Patients attributed to provider	0.4%	0.0001	<0.001	0.2%	0.5%
+10 Providers attributed to Practice	0.1%	0.00031	0.642	−0.5%	0.8%
+10 Patients attributed to Practice	0.0%	0.00	<0.001	−0.1%	0.1%
Provider Specialty (Ref=Infectious Disease)
Generalist	1.6%	0.00482	<0.001	0.6%	2.5%
Other Specialty	5.1%	0.0052	<0.001	4.1%	6.1%
Patient Characteristics
Female	−1.1%	0.0027	<0.001	−1.6%	−0.6%
Race/Ethnicity (ref=white)
Black	−4.5%	0.004045	<0.001	−5.3%	−3.7%
Hispanic	−1.7%	0.004375	<0.001	−2.5%	−0.8%
Asian	1.4%	0.0090	0.125	−0.4%	3.1%
Other	−2.8%	0.0084	0.001	−4.4%	−1.1%
Age	0.5%	0.0001	<0.001	0.4%	0.5%
Years with same provider	6.0%	0.0016	<0.001	5.7%	6.3%
Year (ref=2009)
2010	1.0%	0.0032	0.002	0.4%	1.6%
2011	2.4%	0.0043	<0.001	1.6%	3.3%
2012	0.5%	0.0044	0.28	−0.4%	1.3%
State (ref=FL)
GA	−3.1%	0.0100	0.002	−5.1%	−1.1%
IL	2.5%	0.0094	0.009	0.6%	4.3%
LA	−4.5%	0.0139	0.001	−7.2%	−1.8%
MA	−1.9%	0.0113	0.093	−4.1%	0.3%
MD	−6.7%	0.0157	<0.001	−9.7%	−3.6%
NC	−0.4%	0.0099	0.677	−2.4%	1.5%
NJ	−3.2%	0.0116	0.006	−5.5%	−0.9%
NY	3.4%	0.0063	<0.001	2.2%	4.7%
OH	−3.8%	0.0263	0.15	−9.0%	1.4%
PA	−0.6%	0.0147	0.689	−3.5%	2.3%
TX	−8.1%	0.0097	<0.001	−10.0%	−6.2%
VA	−6.9%	0.0131	<0.001	−9.4%	−4.3%
Count of Chronic Conditions	0.2%	0.0005	<0.001	0.1%	0.3%
Disability as Basis of Eligibility	3.1%	0.0029	<0.001	2.5%	3.6%
Any Drug Abuse Diagnosis	−1.2%	0.0031	<0.001	−1.8%	−0.6%
Any Alcohol Abuse Diagnosis	−1.2%	0.0034	<0.001	−1.9%	−0.6%
Regimen (ref=NRTI based)
NNRTI based	13.2%	0.0072	<0.001	11.8%	14.6%
Boosted PI based	11.4%	0.0071	<0.001	10.0%	12.8%
Integrase based	9.1%	0.0080	<0.001	7.5%	10.7%
PI based	8.9%	0.0076	<0.001	7.4%	10.3%
multiple/others	10.3%	0.0075	<0.001	8.8%	11.8%
Constant	15.4%	0.0137	<0.001	12.7%	18.1%

Notes: All results from a linear hierarchical model with provider random effects nested within practice random effects.

At the patient-level, the factor most associated with adherence was number of years the patient was attributed to the same provider. A one-year increase in years attributed to the same provider was associated with a 6.0 percentage point increase in percent of year adherent (95% CI: 5.7–6.3). Substance use was associated with a lower adherence (−1.2% for drug or alcohol use diagnoses). Black patients and Hispanic patients had lower adherence rates than white, non-Hispanic patients: 4.5 percentage points (95% CI: 3.7–5.3) and 1.7 percentage points (95% CI: 0.8–2.5 percentage points) lower, respectively.

Variance components analysis

Table 3 presents the amount of variance attributable to each level of the model. Providers alone explained 13% of the variance, while practices alone explained 10.4% of the variance. Providers and practices together explained 11.9% of the variance. In the full model that included patient characteristics and random effects, providers and practices accounted for 6.8% of all variation in adherence. Patient random effects explained 452% of the total variance in adherence in this full model.

Table 3:

Variance in ART Adherence attributable to patient, provider, and practice effects

	Random Effect	SD	Variance	% of Total
Model 1: Provider Random Effects	Provider	10.4%	0.011	13.0%
	Residual	26.9%	0.072	87.0%
Model 2: Practice Random Effects	Practice	9.3%	0.009	10.4%
	Residual	27.3%	0.074	89.6%
Model 3: Provider RE+ Practice RE+ Provider and Practice Characteristics	Provider	6.6%	0.004	5.3%
	Practice	7.4%	0.005	6.6%
	Residual	26.9%	0.073	88.1%
Model 4: Provider RE+ Practice RE+ Patient FE + Provider, Practice and Patient Characteristics	Provider	5%	.003	4.1%
	Practice	4%	.002	2.7%
	Patient	18%	.033	45.2%
	Residual	19%	.035	47.9%

Notes: Each subsequent model added new random effects and patient characteristics to characterize variance attributable to provider, practice, or patient characteristics. The residual variance is variation that is not explained by any variables in the model. Model one can be interpreted as the provider that patients are attributed to, explain 13% of the total variance in patient adherence.

All results were largely robust to changes in model specifications and the inclusion of zip code fixed effects, as shown in the Appendix.

DISCUSSION

This study has two main findings. First, even after adjustment for patient characteristics, variance in ART adherence explained by providers and practices is substantial. Second, in multi-level models of patient-level ART adherence, higher adherence levels were observed for patients seeing generalists and other specialists compared with infectious diseases specialists, MD/DO providers compared with non-MDs, providers with more patients, and for patients who had been with the same provider for longer.

This is the first analysis that we are aware of that has quantitated the impact of provider and practice-level effects on patient-level ART adherence. Measuring these effects is difficult, because it requires the attribution of patients to providers and providers to practices so as to create valid “levels” to use in multi-level models. The approach that we developed and implemented for this study can be used to study a variety of provider and practice-level effects across diseases, as long as there is a clinically sensible way to link patients to providers, which in this case was the provision of an ART prescription.

It is widely appreciated that medication adherence is a complex behavior with multiple drivers [24–27]. There is abundant literature on the effectiveness of adherence interventions [26,28,29], with generally similar conclusions – it is very difficult to improve medication adherence, and most interventions have small to no effects. Given the complexity of medication adherence as a behavior, it is no surprise that our most encompassing model captures only 52.9% of the total variance in the adherence outcome variable after adjustments for observed patients, providers, and practice characteristics. More important for our purposes than the total variance explained, is the way in which this explained variance is partitioned among patients, providers, and practices. Most (87%) of the explained variance is explained by patients’ latent traits, but 13% were explained by provider and practice latent characteristics. Thus, the contributions of provider and practice characteristics to patient-level medication adherence can be substantial

There is no recent evidence that directly compares the quality of care provided by HIV providers with different training backgrounds. Our data show that the patients of generalists and “other specialists” had better adherence than patients of infectious diseases specialists. As previously noted, data from early in the ART era suggested that generalist HIV providers with sufficient experience (as measured by HIV care volumes) perform as well as infectious disease trained physicians [6–9]. As PLWH age, they accumulate other chronic conditions [15,30–32]. It may be that patients with HIV who have providers that are comfortable providing care for these other conditions as well as their HIV (“one stop shopping”) are better able to get coordinated primary care, and that this is reflected in better ART adherence [33]. By analogy, it may be that “other specialists” who also provide ART are able to provide more comprehensive care than infectious disease trained doctors. Our data cannot directly confirm this hypothesis. It is also possible that Medicaid patients cared for by infectious disease trained providers are sicker or more complex in ways that our data cannot capture; however, our findings are adjusted for chronic diseases and robust to the inclusion of zip code fixed effects.

We also found that providers with MD and DO degrees had patients with slightly better ART adherence (1.6%) than other providers (NP/PA/other). Earlier work showed that non-physicians who were experienced with HIV and supported by MD trained providers provide care that is equal to that of physicians [8], and we do not believe that the findings presented here are important counter to this earlier work. More broadly, workforce shortages of providers with expertise in HIV care, particularly in rural areas, suggest that in some clinical situations, care led by non-physician providers is appropriate.

We were not surprised by the findings that provider (but not practice) volume of Medicaid PLWH was significantly associated with better ART adherence. The idea that “practice makes perfect” has a long history in health care [34,35], and literature shows that experience with HIV is associated with both better performance on processes of care and with better health outcomes [6,7,36]. We also were not surprised to find that longer continuity of care is associated with better ART adherence. There is a large literature, both in health care more broadly, but also in HIV care, testifying to the importance of provider continuity [37,38].

There are likely characteristics of providers and practices outside of what we can include in our analysis that matter for improving patient care. Structural aspects of care, including the organization of practices, differential follow-up procedures, how providers communicate with patients, providing training, and others, may all have a role in influencing ART adherence. Our findings highlight that providers and practices play an important role in the successes or failures of their patient’s attribution. More research is needed to further disentangle which structural characteristics have the greatest effect on ART adherence and how to change them.

This study has several limitations. First, our design is cross sectional, so our results do not formally have a causal interpretation. There may be unmeasured confounding of the types of patients who go to different providers which we cannot account for in our results. For instance, if more complex patients who have greater adherence challenges require greater specialty care, that could explain some of our findings. Second, our attribution methods may be imperfect, particularly for patients who see many different providers within a year, or if team based care occurs within practices. Further, the practice identifier may be under-detecting providers that provide care at certain locations. If provider or practice attribution is misclassified, then it may draw the variance partition coefficients closer to the null. Third, ideally we would use better defined measures of adherence, such as persistence and implementation [39]. However, our attribution method assigned a provider to a patient at the level of a calendar year, and because of this we believe that the percent adherence measure that we used was most appropriate. Fourth, we are limited in the provider and practice characteristics available in our models, and the variables we currently include may not account for other important factors. Fifth, we assume that most of the providers and practices in our study care for patients with multiple types of insurance, which means that we are likely capturing only a fraction (those with Medicaid) of the persons using ART cared for by the providers and practices identified. However, about 40% of national PLWH are enrolled in Medicaid. Finally, our findings may not generalize to non-Medicaid patients or Medicaid Managed Care patients.

In conclusion, this paper presents a new method to identify treating providers and medical practices using Medicaid claims data. We supplemented this method with analysis that relies on multi-level models. This methodology should be applicable to other clinical conditions. Substantial variance in ART adherence was attributable to variation among clinicians and practices, which suggests that providers and practices can be important targets for adherence interventions. The finding that the patients of generalists and other specialists have better ART adherence than infectious diseases trained providers highlights the need for better care coordination as PLWH have more comorbid conditions and are taking more medications to treat these conditions. These challenges will only increase as PLWH age.