The impact of HCV co-infection status on healthcare-related utilization among people living with HIV in British Columbia, Canada: a retrospective cohort study

The province of BC established the BC Centre for Excellence in HIV/AIDS Drug Treatment Program (DTP) in 1992; it has since been responsible for the distribution of antiretrovirals [22]. The DTP, funded by the provincial government, provides HIV medical care and laboratory monitoring (e.g., CD4 cell counts and viral load) for all diagnosed PLWH residing in BC at no cost, in accordance with BC’s HIV therapeutic guidelines, which have largely remained consistent with those of the International Antiviral Society-USA since 1996 [23, 24].

This retrospective study was carried out using data from the British Columbia Seek and Treat for Optimal Prevention of HIV/AIDS (STOP HIV/AIDS) population-based cohort, which is derived from various linkages among provincial databases [22, 25‐30]. This cohort is briefly described in the Additional file 1. Our inclusion criteria for individuals were as follows: (i) ART-naïve individuals aged ≥18 years, (ii) enrolled in the (DTP) between January 1, 2000 and December 31, 2013, (iii) initiated ART consisting of two nucleoside/nucleotide reverse transcriptase inhibitors (NRTIs) as backbone, plus either a non-nucleoside reverse transcriptase inhibitor (NNRTI) or a ritonavir-boosted protease inhibitor (bPI), (iv) had a CD4 count and a viral load measurement within 6 months of ART treatment initiation, and (v) had at least 6 months of follow-up. Note that we decided to exclude other initial ART regimens due to the small number of individuals who have initiated on them. Eligible individuals were followed until December 31, 2014, the last contact date (i.e., the last available laboratory test date, the last filled ART prescription refill date or the date of last encounter with the healthcare system identified in any of the STOP HIV/AIDS databases), or the date of death (all-causes).

All viral load tests and the majority of CD4 cell count tests were performed by the St. Paul’s Hospital laboratories in Vancouver, BC, and were subsequently transferred to the DTP via electronic linkage. CD4 cell count tests completed at other laboratories throughout BC were manually entered into the DTP; altogether resulting in approximately 85% data capture of all CD4 cell count tests done in the province. For analytical purposes, all viral load measurements were transformed to range from < 50 (coded as 49) to > 100,000 (coded as 100,010) copies/mL. This process was necessary to account for advances in testing methodology, as previously described elsewhere [31]. CD4 cell counts were measured by flow cytometry (Beckman Coulter, Inc., Mississauga, Ontario).

The main outcome was the rate of HRV per individual for every 6-month interval. HRV was calculated based on records from the Medical Services Plan (MSP) billing provincial database linked to STOP HIV/AIDS cohort. MSP captures HIV and non-HIV-related inpatient and outpatient services provided by physicians and supplementary health care practitioners, as well as diagnostic procedures. The unit of analysis for the crude HRV rate was person-year, which was calculated by dividing the number of HRV by the number of person-years of follow-up in each calendar year. The corresponding 95% confidence intervals (CI) for these rates were based on the Fisher’s exact test [32]. Note that we also presented the age-sex standardized HRV rates for the overall follow-up utilizing BC’s population estimates as the reference [33].

The number of HRV was derived from the MSP physician billing records, which contain the date, type and location of service, physician number, practitioner speciality and costs. Since multiple MSP database records could be associated with one unique HRV, a record was considered to be a unique HRV if it satisfied one of the following conditions:

Once the rules outlined above were adopted, we identified the unique HRV for these individuals. Subsequently, we calculated the number of HRV in each 6-month interval for all individuals. Visits were classified as general practitioners, other healthcare practitioners, and laboratory services.

The main exposure variable of interest was HIV/HCV co-infection status, derived from the DTP database, which indicates whether PLW-HIV has ever had evidence of HCV infection (i.e., a HCV antibody positive or HCV RNA detected, as indicated by laboratory result data or physician reported status). Note that upon successful HCV treatment, individuals can become re-infected with HCV. Thus, we were only able to incorporate HCV ever status for these individuals.

The confounders measured at ART initiation included: sex (male, female), risk for HIV acquisition (gay, bisexual and other men who have sex with men (MSM), PWID, MSM/PWID, Other, Unknown), initial ART regimen (NNRTI, bPI) and period of ART initiation (2000–2003, 2004–2007, 2008–2011, 2012–2013). Several time-varying confounders, measured every year, were considered in this study, including: age (< 30, 30–39, 40–49, ≥50 years), CD4 cell count (< 50, 50–199, 200–349, ≥350 cells/mm³), viral load (log₁₀ transformed), adherence level (< 40%, 40–79%, 80–94%, ≥95%), cumulative number of comorbid diseases (0, 1, 2, ≥3) and person-years of follow-up time. Adherence level was determined on the basis of a validated measure assessing refill compliance, which was calculated by dividing the amount of days of dispensed ART medication by the amount of days of study follow-up, for each period (presented as percentage). ART adherence calculations were derived from distinct regimen exposures for each individual [34]. PWID included individuals with past and current exposure to injection drug use. Individual comorbidities were derived from the Charlson Comorbidity Index (Additional file 1: Table S1) [35]. These included 16 conditions, other than HIV/AIDS (e.g., renal, liver, heart and lung diseases and cancer). These conditions were identified using International Classification of Diseases (Ninth and Tenth Revisions, Clinical Modification) diagnosis codes obtained from the STOP HIV/AIDS cohort databases based on a validated case-finding algorithm [36].

Categorical variables were compared using the Fisher’s exact test or the X² test, and continuous variables were compared using the Kruskal-Wallis test [37]. Based on exploratory data analyses to determine the best distribution for modeling HRV rates, non-linear mixed effects models were used assuming a Poisson distribution, the person-years of follow-up time as the offset, a log link function, a random intercept term and an autoregressive of order one working correlation matrix [38]. We have chosen these models since they are flexible in taking into account the inter- and intra-individual sources of variation, they can handle imbalanced longitudinal data, and zero-inflated models did not show any gain over the final fitted model [38, 39]. Potential confounders were selected for inclusion in the final model using a backward-selection approach, published by our group based on the work by Maldonado and Greenland [40], that considered the magnitude of change in the coefficient of the HIV/HCV co-infection status variable. Specifically, starting with a fixed model, which considered all available variables, potential confounders were dropped one at a time, using the relative change in the coefficient for the variable related to the HIV/HCV co-infection status as a criterion, until the maximum change from the full model exceeded 5% [41]. Note that in the multivariable model, we did not adjust for risk for HIV acquisition given its high collinearity with the main study exposure. All analyses were performed using either R© version 3.3.2 (The R foundation for statistical computing, Vienna, Austria) or SAS version 9.4 (SAS, Cary, North Carolina, USA).

Overall, 4217 ART-naïve adults with a total of 615,776 HRV were initially eligible to participate in this study. Among these individuals, 81% were males, 67% were aged between 30 and 49 years, 72% were either MSM, PWID or both, 40% started ART between 2008 and 2011, 76% initiated ART with a CD4 cell count < 350 cells/mm³ and had a median viral load 4.90 log₁₀ copies/mL (25th–75th percentile (Q1-Q3): 4.38–5.00 log₁₀ copies/mL), 52% started on a bPI-based ART, 78% had adherence ≥95% during the first six months on ART, and 34% had no comorbidities while 14% had 3 or more. The median follow-up time was 4.99 (Q1-Q3: 2.50–7.98) years, in which the median number of HRV was 97 (Q1-Q3: 48–187) (Table 1). This cohort comprised of 2333 (55%) PLW-HIV, 1622 (39%) PLW-HIV/HCV, and 262 (6%) whose HCV status was unknown. The distribution of study variables among those included in this analysis (N = 3955) was very similar to the original 4217 individuals described above. For the purposes of this study, those with unknown HCV status, who contributed 25,393 HRV (4% of total visits), were excluded from the subsequent analyses. As noted in Table 1, those excluded were more likely to have risk for HIV acquisition other than MSM or PWID, a lower number of HRV and a slight shorter follow-up time.

Bivariable analysis exploring associations between study characteristics and HCV co-infection status (shown in Table 2) revealed that PLW-HIV and PLW-HIV/HCV differed significantly in all study characteristics, except for initial ART regimen. PLW-HIV/HCV were more likely to be younger, female, PWID, initiate ART prior to 2008, have a lower CD4 cell count and viral load measurement at baseline, and maintain adherence < 40% during follow-up. Additionally, these individuals were also more likely to present with and develop a higher number of comorbidities compared to PLW-HIV. The number of PLW-HIV/HCV with moderate or severe liver disease increased substantially during follow-up (75 (5%) to 124 (8%); p-value 0.0004). Apart from liver disease, chronic pulmonary disease and cancer (all causes) were the most prevalent comorbidities (results not shown).

The overall age-sex standardized HRV rate for PLW-HIV and PLW-HIV/HCV was 21.11 per person-year (95% CI 20.96–21.25) and 41.69 per person-year (95% CI 41.51–41.88), respectively. For PLW-HIV/HCV, visits were mainly related to methadone or buprenorphine/naloxone treatment, and not to the same extent, to laboratory tests done to monitor liver and kidney function and the hematology profile. For PLW-HIV, most visits were concerned with the latter. As demonstrated in Fig. 1a, the annual crude HRV rates in both groups have been steadily decreasing from 2000 to 2013. The decrease in HRV rates was more prominent for PLW-HIV (34.28 per person-year to 17.98 per person-year; 48% decrease; p-value < 0.0001) compared to PLW-HIV/HCV (48.57 per person-year to 33.92 per person-year; 30% decrease; p-value < 0.0001). The trends in HRV rates stratified by HIV/HCV co-infection status and by the type of HRV are illustrated in Fig. 1b. As shown, except for HRV rates related to laboratory services, PLW-HIV/HCV consistently maintained higher rates, especially those related to general practitioner visits (> 2 times higher). Note that the rates for other healthcare practitioners (which includes visits to specialists), although much lower than the HRV rates for general practitioner visits, they have been more stable over time and the difference between these groups was not as pronounced. More detailed information on these trends can be found in Additional file 1: Table S2. The result of the multivariable confounder model showed that the adjusted HRV rate ratio for PLW-HIV/HCV in comparison to PLW-HIV was 1.18 (95% CI 1.13–1.24), after controlling for sex, age, ART era, time-varying CD4, adherence to ART, viral load and the number of comorbidities (Table 3).