Health Care Costs in a Cohort of HIV-Infected U.S. Veterans Receiving Regimens Containing Tenofovir Disoproxil Fumarate/Emtricitabine

BACKGROUND: Tenofovir disoproxil fumarate (TDF), a key component in many human immunodeficiency virus (HIV) treatment regimens, is associated with increased renal and bone toxicities. The contributions of such toxicities to treatment costs, as well as the relative differences in treatment costs for various TDF/emtricitabine (FTC) regimens, remains unexplored. OBJECTIVE: To estimate and compare mean overall and renal- and bone-specific costs, including total, inpatient, outpatient, and pharmacy costs in patients treated with TDF/FTC+efavirenz (EFV) compared with several non-EFV-containing TDF/FTC regimens. METHODS: We conducted a national cohort study of treatment-naive HIV-infected U.S. veterans who initiated treatment from 2003 to 2015 with TDF/FTC in combination with EFV, elvitegravir/cobicistat, rilpivirine, or ritonavir-boosted protease inhibitors (atazanavir, darunavir, or lopinavir). Outcomes of interest were quarterly total, inpatient, outpatient, and pharmacy costs using data from the Veterans Health Administration (VHA) electronic medical record and Managerial Cost Accounting System (an activity-based accounting system that allocates VHA expenditures to patient encounters). We controlled for measured confounders using inverse probability of treatment (IPT) weights and assessed differences using standardized mean differences (SMDs). For comparisons where SMDs exceeded 0.1 after IPT weighting, we used the more conservative matching weights in sensitivity analyses. For hypothesis testing, we compared IPT-adjusted differences in quarterly costs between treatment groups using Mann-Whitney U-tests and generalized estimating equation (GEE) regression models. RESULTS: Of 33,048 HIV-positive veterans, 7,222 met eligibility criteria, including 4,172 TDF/FTC + EFV recipients; mean (SD) age of the cohort was 50.0 (10.0) years; 96.7% were male; 60.1% were black; and 30.1% were white. Quarterly periods of exposure to EFV-containing regimens were 22,499 and of exposure to non-EFV-containing regimens were 11,633. After IPT weighting, absolute SMDs were < 0.1 except for a few covariates in the rilpivirine comparison. The per-patient adjusted mean total quarterly costs were $7,145 for EFV versus $8,726 for non-EFV (P < 0.001; Mann-Whitney U-test) and the per-patient adjusted mean difference in total quarterly costs was $1,419 lower for EFV versus all non-EFV combined (P < 0.001; GEE model). Corresponding values for outpatient costs ($2,656 vs. $2,942; P < 0.001; difference, -$254; P = 0.001), inpatient costs ($2,009 vs. $2,614; P < 0.001), radiology costs ($213 vs. $276; P < 0.001), and pharmacy costs ($2,480 vs. $3,170; P < 0.001; difference, -$600; P < 0.001) were all lower for EFV versus all non-EFV combined. Findings based on matching weights were qualitatively similar. Contributions of renal and bone costs to the total costs of treatment were very small, ranging between $52 and $94 per patient per quarter for renal outcomes and between $6 and $114 for bone outcomes. CONCLUSIONS: Among 7,222 HIV-treated veterans over an average follow-up of 1.2 years per patient, those patients receiving TDF/FTC + EFV had lower overall health care costs compared with those receiving non-EFV regimens.

T enofovir disoproxil fumarate (TDF)/emtricitabine (FTC) is the most commonly used antiretroviral therapy (ART) backbone for treating human immunodeficiency virus (HIV) worldwide. 1 However, epidemiologic studies have shown increased renal and bone toxicities associated with TDF, contributing to guideline recommendations that TDF should be used with caution or avoided in patients with renal disease and osteoporosis. 1 However, switching patients who are currently stable on TDF/FTC-based regimens, which continue to be recommended in World Health Organization guidelines, 2 should be approached cautiously, particularly as newer evidence has identified that the renal and bone adverse effects may be attenuated by pairing TDF/FTC with certain third agents. For example, TDF/FTC regimens that include efavirenz (EFV) have been associated with lower rates of adverse renal and bone outcomes compared with the addition of other third agents. EFV, a non-nucleoside reverse transcriptase inhibitor (NNRTI), has been associated with lower risks of both proteinuria and chronic kidney disease (CKD) compared with tenofovir, a nucleoside reverse transcriptase inhibitor 3 ; lower TDF-associated renal impairment in ART combinations that included EFV compared with nevirapine, an NNRTI 4 ; and preserved bone mineral density (BMD) compared with atazanavir (ATV), a protease inhibitor (PI). [5][6][7][8] Exclusion Criteria. Patients were excluded if they had a pharmacy claim for any ART agent in the pre-index period, defined as the 6-month period before the index date. To further ensure each patient was antiretroviral naive and not receiving continuing ART from previous treatment outside the VHA, we used a validated method for excluding patients who had most likely received previous treatment. 20 This method excluded patients if (a) the index regimen was a "salvage" regimen (i.e., composed of both a PI and NNRTI or composed of ≥ 5 agents) or (b) any HIV RNA level < 500 copies per mL during the preindex period, which suggested prior antiretroviral exposure. 20 We also excluded patients who did not have an inpatient or outpatient encounter at least 6 months before the index date to ensure that we captured patients with a pattern of receiving routine care within the VHA.
Follow-up. Patients were followed until they discontinued the ART regimen of interest or until the end of the study period (December 31, 2015), whichever was sooner. We defined discontinuation as a gap of more than 30 days for any 1 of the medications in a patient's ART regimen. The date of discontinuation was the first day of such a gap.
Estimates of the real-world economic burden of TDF/FTCassociated adverse events (AEs), including renal and bone outcomes, are lacking in the existing literature. Nadkarni et al. (2015) reported that the incidence of hospitalizations for acute kidney injury in patients with HIV doubled from 2002 to 2010, which they attributed to the aging of the HIV population and increases in comorbidities; however, ART effects on acute kidney injury were not examined. 9 Magoni et al. (2011), who reported the effect of HIV on renal disease in the Italian health care system, also did not evaluate ART effects. 10 Koenig et al. (2010) evaluated the cost-effectiveness of monitoring for renal insufficiency in Haitian HIV patients on ART and found it to be cost prohibitive with negligible clinical benefits in a resource-limited setting. 11 In a cost-effectiveness modeling study, Walensky et al. (2016) determined that the reduced renal and bone toxicity of tenofovir alafenamide justified up to a $1,000/year increase in cost for tenofovir alafenamide versus TDF, 12 largely due to an assumed high cost for hemodialysis of $87,000 based upon an estimated rate of hemodialysis of 25% in those discontinuing TDF. However, this assumption is not supported by evidence from the Data collection on Adverse events of Anti-HIV Drugs (D:A:D) study demonstrating that end-stage renal disease (ESRD) in patients with HIV infection over a median follow-up of 6 years was associated with diabetes, hypertension, baseline estimated glomerular filtration rate (eGFR), smoking, and current CD4 cell count rather than past or current exposure to TDF. 13,14 The real-world cost differences associated with treatment and maintenance of HIV patients across TDF/FTC regimens in combination with different third agents in a U.S. setting remain unknown. The relative contributions of renal and bone adverse outcomes to these costs are also unknown, both on a per-patient basis and on a health-system basis. Given that TDF/ FTC regimens that include EFV have demonstrated relatively lower rates of adverse renal and bone outcomes compared with other TDF/FTC regimens, 3,4,6-8 a reduced risk of these adverse outcomes with EFV may also be associated with lower health care costs with EFV compared to other third agents.
Our primary objectives in this analysis were to (a) estimate the adjusted mean total, outpatient, and pharmacy costs of HIV treatment with TDF/FTC in combination with several common base agents and (b) compare adjusted mean differences in total, outpatient, and pharmacy costs. We did this in a cohort of treatment-naive U.S. veterans initiating treatment with TDF/FTC plus EFV, elvitegravir/cobicistat (EVG/c), rilpivirine (RPV), or boosted PIs. In a supplemental analysis, we also examined the renal-and bone-specific components of costs and use of the Veterans Affairs (VA) health care system.

Study Variables
Outcomes/Dependent Variables. Our primary outcome variables included quarterly total, outpatient, inpatient, and pharmacy costs during the period following the index date up to discontinuation of the regimen of interest or the end of the study period. These quarterly costs were calculated by summing all VHA health care costs for inpatient, outpatient, and pharmacy services provided in each quarter after the index date based on data available via the Managerial Cost Accounting System (an activity-based accounting system that allocates VHA expenditures to patient encounters). Quarterly use outcomes were assessed by counting frequencies of outpatient visits and numbers of inpatient days for hospital stays in the VHA during the quarter in which they occurred.
To contrast with the overall costs and use previously described, in supplemental analyses, we calculated quarterly bone-and renal-specific costs and use by summing all VHA health care costs and use associated with an ICD-9/10-CM or Current Procedural Terminology, 4th Edition (CPT-4) code for bone disease (i.e., osteoporosis diagnosis or a vertebral, hip, or wrist fracture) or renal disease (i.e., proteinuria, CKD, or ESRD, including renal transplant and dialysis). All codes for bone and renal disease are available via Web links 1 and 2, respectively.
We used cost per patient-quarter as our unit of measure because duration of follow-up varied substantially between individuals. All cost outcomes were converted to 2015 U.S. dollars (USD) using the personal consumption expenditures index published by the U.S. Bureau of Economic Analysis. 24 Exposures/Independent Variables. The exposure of interest was TDF/FTC in combination with EFV compared with TDF/ FTC in combination with (a) RPV, (b) EVG/c, or (c) boosted PIs. These 3 TDF/FTC-based FDC regimens were ascertained based on pharmacy claims information.
Covariates/Control Variables. Patient characteristics with a known or theoretical association with exposures and outcomes were identified based on literature search and expert consensus. Identified covariates measured during the pre-index period included demographics (i.e., age, race, body mass index, marital status, sex), disease severity measures (i.e., CD4 count, viral load), measures of comorbid disease (i.e., bone disease, albumin, eGFR, and comorbid conditions), drug exposures (i.e., methadone, proton pump inhibitors, bisphosphonates), and cost. These were assessed in the 6 months before each patient's index date and were used to control for potential confounding.
Bone disease during the pre-index period was defined as a diagnosis of osteoporosis/osteopenia by ICD-9/10-CM diagnosis codes, CPT-4 codes, or classification by BMD test result. Data on patient BMD T-scores were derived from narrative records using a validated natural language processing (NLP) technique. 21 NLP can be used to extract clinical data from radiology reports, clinicians' progress notes, and other text reports. This validated technique has been applied in previous studies to extract BMD T-scores from radiology reports. [21][22][23] In validation, an examination of test characteristics for the tool revealed that the NLP tool for BMD from radiology reports had an overall accuracy of 92.8% for the anatomic site at which the BMD was measured, 92.8% for the value of the BMD T-score, and 90.4% for the matching the correct BMD T-score value with the correct anatomical site. 21

Statistical Analysis
Outcomes. We reported patient characteristics during the pre-index period using descriptive statistics and assessed differences between the groups using standardized mean differences (SMDs); absolute SMDs >0.1 were considered meaningful differences. 25 To control for confounding by indication and selection bias, we estimated inverse probability of treatment (IPT) weights, 26 an extension of the propensity score, which represents the conditional probability of receiving the treatment as predicted by covariates described above. IPT weights were calculated from a propensity score generated using logistic regression that estimated the probability that each patient would initiate TDF/FTC plus EFV. 26 In the event that covariate balance using IPT weights was not achieved for any of the comparisons, we conducted a sensitivity analysis using matching weights. 27 For our multivariable models, we included a category for "missing" data. 28 We compared adjusted costs and use outcomes using IPTweighted Mann-Whitney U-tests to compare distributions for all pair-wise comparisons versus EFV. We also conducted IPT-weighted generalized estimating equation (GEE) regression models to estimate the adjusted mean differences in quarterly overall outpatient, pharmacy, and total costs and outpatient use associated with exposure to each regimen of interest in the weighted cohort. We identified the most appropriate distribution for the GEE model by using the modified Park test. 29 In addition, we used the method of recycled predictions to generate estimates of the marginal effect of each exposure on cost and use in the original scale. 30 The relative change in cost associated with EFV-based regimens versus each non-EFV regimen was expressed as the expenditure rate ratio (ERR).

Supplemental Analysis.
For adjusted mean differences of overall inpatient and radiology costs and use, as well as for adjusted mean differences of any renal and bone cost or use, we used a 2-part model 31 to calculate the marginal effect associated with each exposure, conditional upon having that cost or use. Specifically, part 1 of the model assessed the odds of having a positive cost (logistic regression) and part 2 assessed the ERR for EFV versus other regimens conditional on having a positive cost (IPT-weighted GEE).

Health Care Costs in a Cohort of HIV-Infected U.S. Veterans Receiving Regimens Containing Tenofovir Disoproxil Fumarate/Emtricitabine
After IPT weighting, all absolute SMDs were < 0.10 except for 5 covariates in the RPV comparison, which were between 0.1 and 0.2; these included age, viral load, CKD, cardiovascular disease, and diabetes ( Figure 2). After weighting with matching weights, all SMDs were < 0.1 (data not shown), meaning the results were qualitatively similar.

Outcomes
Adjusted Mean Total, Inpatient, Outpatient, and Pharmacy Costs. Adjusted mean overall costs were significantly lower for EFV versus all other regimens for every type of cost, including total, outpatient, inpatient, radiology, and pharmacy (all P < 0.001; Table 2). Adjusted mean total quarterly cost associated with EFV was $7,145, compared with $8,532 with PI-based regimens (the most commonly prescribed drug class of the non-EFV regimens), $8,726 with all non-EFV regimens combined, $11,258 with RPV-based regimens, and $11,728 with EVG/c-based regimens. In the GEE regression, the adjusted mean differences of total quarterly costs were $1,419 (P < 0.001), $4,488 (P = 0.012), $3,372 (P = 0.042), and $1,303 (P < 0.001) lower for EFV regimens versus non-EFV overall, EVG/c, RPV, and PI regimens, respectively (Table 3). Total costs and pharmacy costs were consistently statistically significant, with pharmacy costs being most greatly reduced for EFV in comparison to EVG/c (a reduction of $1,876; P < 0.001). For EFV regimens versus non-EFV regimens, outpatient costs were $254 (P = 0.001) lower, while pharmacy costs were $600 lower (P < 0.001). Compared with all other regimens of interest, ERRs indicated that EFV regimens were associated with significantly lower quarterly expenditure rates for outpatient (9.0%-17.4% lower), pharmacy (20.4%-46.1% lower), or total costs (15.7%-39.5% lower; Table 3). Unadjusted means and standard deviations are presented in Appendix A (available in online article).   (Table 3).
Notably, renal and bone use followed the trends of overall use, being generally lower for patients on EFV than on any other regimen and significantly so only in comparisons with EVG/c and RPV regimens ( Table 2).

Supplemental 2-Part Models.
Because very few patients had zero outpatient or pharmacy costs in the overall analyses, 2-part models were not undertaken for these outcomes in that analysis. In overall 2-part models, the odds of having a positive inpatient or radiology cost were significantly lower for patients on EFV compared with those on PI-based regimens (part 1 models: Appendix B, available in online article). However, among patients who incurred an overall inpatient or radiology cost, none of the mean differences between regimens for such costs Adjusted Mean Renal-and Bone-Specific Costs. Adjusted mean renal and bone costs were only a small fraction of the overall costs associated with each exposure. Quarterly renal costs ranged from $29 on the RPV regimen (vs. $11,258 total costs on RPV) to $88 on the EVG/c regimen (vs. $11,728 total costs on EVG/c). Bone costs ranged from $6 on the EVG/c regimen (vs. $11,728 total costs) to $114 on the RPV regimen (vs. $11,258 total costs). Differences were statistically significant for many comparisons, with some comparisons favoring EFV and others favoring comparators (Table 2).

Inpatient and Outpatient Utilization.
Overall outpatient and inpatient use on EFV regimens were also significantly lower compared with all non-EFV regimens as well as each of the individual regimens ( Table 2). The overall total quarterly outpatient encounters were lower with EFV compared with all non-EFV regimens; mean differences were −2.39 (P < 0.001),     Similarly, the odds of having positive inpatient use were lower with EFV versus all non-EFV regimens; however, among these patients with positive inpatient use, there were no significant adjusted mean differences in inpatient use (Appendix B). In renal 2-part models, the odds of having a positive pharmacy cost was significantly higher versus EVG/c; however, among these patients with a positive cost, the adjusted mean difference in cost was not significant (Appendix B). The odds of having positive inpatient costs, inpatient use, and outpatient use were significantly lower with EFV versus a number of comparators; however, among these patients with positive costs, significant adjusted mean differences in cost were found for only 2 comparisons (EFV vs. PI for inpatient cost, difference $10,254, P = 0.041; and EFV vs. RPV for outpatient encounters, difference 0.69, P = 0.014; Appendix B).

Patient Characteristics in the Pre-Index Period: Verification that IPT Weighting Achieves Covariate Balance Between TDF/FTC-Containing Regimens with EFV Versus All Non-EFV, EVG/c, RPV, and RTV-Boosted PIs
In bone 2-part models, the odds of having a positive inpatient cost was significantly lower versus RPV; however, among these patients with a positive cost, the adjusted mean difference in cost was not significant (Appendix B). No other bone outcomes showed significantly different odds of having a positive cost.

■■ Discussion
We identified the cost and use effects of different third agents in TDF/FTC ART regimens within a nationwide VHA setting. Adjusted mean quarterly overall costs and use consistently and significantly favored EFV over boosted PIs, EVG/c, and RPV in combination with TDF/FTC, including total, inpatient, outpatient, pharmacy, and radiology costs. We found that patients receiving EFV regimens had quarterly costs that were $1,419 (95% CI = $843-$1,994) lower than those receiving all non-EFV   Figure 1) versus non-EFV regimens may be due to several possible reasons. They may be directly related to regimen cost or unmeasured concomitant medications; alternatively, they may be due to unmeasured adverse outcomes from other comorbidities exacerbated by TDF/FTC regimens. Although renal and bone use followed the trends of overall use in the primary analysis, 2-part models, exploring renal and bone costs and use in the smaller population of patients who did have positive costs or use, did not consistently follow the overall results. Renal and bone costs were a very small proportion of overall costs, and the CIs in analysis were often wide in these 2-part analyses due to the small sample sizes. Because renal-and bone-specific costs made up such a small proportion of overall costs (< 1% according to our data), we concluded that renal and bone disease were not major contributors to the overall cost of long-term HIV treatment in this population.
The disconnect between the overall and the renal and bone costs can likely be attributed to the rarity of renal and bone outcomes but also to other, unmeasured health care needs. HIV drug costs are the main cost for long-term HIV patients, although estimates vary from setting to setting. 18,32 The cost-effectiveness analysis (CEA) conducted by Simpson et al. (2013) provides a list of other common health care needs for patients with HIV, 18

Health Care Costs in a Cohort of HIV-Infected U.S. Veterans Receiving
Regimens Containing Tenofovir Disoproxil Fumarate/Emtricitabine differences to be made. Nonetheless, IPT weighting is unable to control for unmeasured confounders that may affect the choice of initial ART or unmeasured outcomes in the study. Second, the costs and use captured in our study likely represent an underestimate of total costs and use within the cohort because of health care encounters and costs being limited to those occurring within the VHA; a quarter to nearly one-half of veterans receive some health care in non-VHA facilities. [38][39][40] Studies comparing demographic and disease characteristics in populations receiving exclusive VHA versus mixed VHA/ Medicare services are conflicting; for example, 1 investigation found that patients considered at highest risk were more likely to be receiving mixed VHA and Medicare than exclusive VHA services, 41 and in another investigation, patients with higher indices of socioeconomic advantage were more likely to be receiving mixed versus exclusive VHA services. 42 However, given the age distribution of our VHA population, less than 5% would have been eligible for Medicare, making it unlikely that dual health care usage affected our conclusions.
Third, because the follow-up time for patients in our cohort varied substantially, we divided the follow-up period into quarters. Alternative statistical methods exist to analyze health care costs in the presence of censoring. For example, Lin et al. (1997) proposed an approach that constructs a survivaladjusted estimator. 43 Bang and Tsiatis (2000) improved on this approach by performing inverse probability weighting with survival from censoring probabilities and assuming continuous death and censoring times. 44 Basu and Manning (2010) extended these methods to allow for estimation of the effect of covariates on survival effects and intensity effects separately. 45 Future work should apply these methods to estimate the health care costs associated with TDF/FTC ART regimens.
Fourth, the VHA patient population has a small number of females, which limits the generalizability of the results to the wider population of patients with HIV infection, particularly concerning osteoporosis, where disease burden and treatment differ substantially between men and women. 46 Finally, because the outcomes have been historically obtained rather than gathered prospectively, the lack of ascertainment of non-VHA pharmacotherapies or care limits the generalizability of these findings to non-VHA contexts.

■■ Conclusions
We found that patients on TDF/FTC regimens with EFV had lower overall health care costs and were less likely to have inpatient admissions than those on regimens of TDF/FTC plus RPV, EVG/c, or boosted PIs.
Medicaid data from the early highly active antiretroviral therapy era (i.e., 2002-2003, all inflated to 2011 USD). For example, chronic opportunistic infections requiring ongoing treatment were found to cost anywhere from $5,052 annually for cytomegalovirus colitis to $120,000 annually for cytomegalovirus retinitis. 18 Nonserious side effects priced low (from $0 to $223 per episode), while serious side effects (including renal failure at $7,913 per episode) and acute opportunistic infections were more costly, ranging from $1,336 for an episode of viral infection causing cervical cancer to $20,248 for a histoplasmosis episode. Although the current relative costs of HIV-related events can only be approximated from this list, due to its age and the fact that it focuses primarily on health care needs of patients on PIs, we can infer the relative frequency, importance, and costliness of renal events compared with other HIV-and ART-related events such as acute and chronic infections, even if it does not provide us with real-world incidences of these needs.
Numerous CEAs that compare various ART regimens, several of which include renal AEs as model inputs, rarely feature renal or bone disease as an outcome associated with ART generally, and none with TDF/FTC specifically. [33][34][35] One CEA by Juday et al. (2013) determined that the FDC TDF/FTC/EVG/c was not cost-effective when compared with the FDC TDF/FTC/ EFV at $100,000/quality-adjusted life years; these results were very sensitive to likelihood of renal AEs. 36 Ultimately, most of these analyses considered renal disease as only a contributing factor to the overall cost of HIV or ART, and only one 12 considered bone disease as a contributing factor. Cumulatively, these studies and other CEAs investigating ART generally [15][16][17][18][19] indicate a research need to analyze the economic burden of various components of HIV care associated with different TDF/ FTC-based ART regimens.
A key strength of this study can be attributed to the use of VHA datasets. The VHA is the largest integrated provider of HIV care in the United States, allowing a large sample size and to evaluate multiple subgroups. 37 The majority market share, single payer, and integrated care attributes of this population enable observations to be made with robust estimates across the HIV continuum of care.

Limitations
Limitations of this study include those typical of observational studies. First, there was an increased risk of residual confounding due to unmeasured differences in patient characteristics during the pre-index period. However, IPT weighting can reliably produce 2 groups that are balanced with respect to measured confounders, allowing valid inferences about treatment