Predictors of unstructured antiretroviral treatment interruption and resumption among HIV-positive individuals in Canada

Abstract

Background

Sustained optimal use of combination antiretroviral treatment (cART) has been shown to decrease morbidity, mortality and HIV transmission. However, incomplete adherence and treatment interruption (TI) remain challenges to the full realization of the promise of cART. We estimated trends and predictors of treatment interruption and resumption among individuals in the Canadian Observational Cohort (CANOC) collaboration.

Methods

cART-naïve individuals ≥18 years of age who initiated cART between 2000–2011 were included. We defined TIs as ≥90 consecutive days off cART. We used descriptive analyses to study TI trends over time and Cox regression to identify factors predicting time to first TI and time to treatment resumption after a first TI.

Results

7,633 participants were eligible, of whom 1,860 (24.5%) experienced a TI. The prevalence of TI in the first calendar year of cART decreased by half over the study period. Our analyses highlighted a higher risk of TI among women (adjusted hazard ratio (aHR): 1.59, 95%CI: 1.33–1.92), younger individuals (aHR: 1.27, 95%CI: 1.15–1.37 per decade increase), earlier treatment initiators (CD4 count ≥350 versus <200 mm³, aHR: 1.46, 95%CI: 1.17–1.81), Aboriginal participants (aHR: 1.67, 95%CI: 1.27–2.20), injecting drug users (aHR: 1.43, 95%CI: 1.09–1.89), and users of zidovudine versus tenofovir in the initial cART regimen (aHR: 2.47, 95%CI: 1.92–3.20). Conversely, factors predicting treatment resumption were male sex, older age, and a CD4 cell count <200 mm³ at cART initiation.

Conclusion

Despite significant improvements in cART since its advent, our results demonstrate that TIs remain relatively prevalent. Strategies to support continuous HIV treatment are needed to maximize the benefits of cART.

Introduction

The expanded use of combination antiretroviral therapy (cART) since 1996 has dramatically enhanced the quality of care and the life expectancy of HIV-positive individuals [1]. However, sustained optimal use of cART is necessary to ensure maximum therapeutic benefits. Incomplete adherence and treatment interruptions (TIs) due to treatment fatigue, side effects and cART toxicities [2–4] have emerged as major challenges to the full realization of the therapeutic promise of cART effectiveness.

Explored as a strategy to reduce cost and cART-related toxicities and improve patient quality of life, TIs, whether physician-directed (structured) or patient-initiated (unstructured), have been found to promote viral rebound and CD4 cell loss, and more importantly, to increase the risk of opportunistic infections and death in observational studies and prospective clinical trials [5–14]. As evidence accumulated demonstrating the adverse effects of TIs, culminating in the publication of results from the seminal SMART trial in 2006, [15] TIs were no longer recommended. However, results from several studies indicate that unstructured, patient-elected TIs continue to occur [5, 16, 17]. Despite the relatively high frequency of TIs, their determinants and outcomes from the year 2000 onwards, a time period often characterized as the era of modern cART [16, 18], and since 2006 when treatment recommendations were developed that precluded physician-directed TIs, are still not well-characterized [16].

As the paradigm of “treatment as prevention” and universal treatment become entrenched in contemporary treatment guidelines in North America [19–22] and globally [23], it is vital that we minimize the occurrence of TIs. In particular, the World Health Organization’s revised guidelines effectively increase the number of individuals eligible for treatment globally by 50% [23], providing an urgent impetus to better understand and address TIs.

Thus, we conducted the present analysis to characterize trends and determinants of treatment interruption and resumption in a Canadian setting of universal free access to HIV care, including medical and laboratory monitoring as well as cART. We hypothesized that individuals who are intermittent or episodic users of cART (defined here as those with gaps in cART of at least 90 days [5, 10, 24]) are different from continuous users. Moreover, we hypothesized that TIs should be less common than in previous studies due to improvements in cART profiles over time. We also assessed characteristics of individuals more likely to reinitiate cART once interrupted.

Methods

Study Population

The Canadian Observational Cohort (CANOC) collaboration is a national collaboration of eight cohorts situated in three provinces (British Columbia (BC), Quebec, and Ontario) of antiretroviral-naive HIV-positive individuals initiating cART after January 1, 2000. The cohort has been described in more detail elsewhere [25]. Briefly, patient eligibility criteria for inclusion into CANOC are documented HIV infection, residence in Canada, at least 18 years of age, initiation of a first antiretroviral regimen comprised of at least three agents, and at least one measurement of HIV plasma viral load and CD4 cell count within six months of initiating cART. Patient selection and data extraction are performed locally at the data centers of the participating cohort studies. Non-nominal data from each cohort on a predefined set of demographic, laboratory, and clinical variables are then pooled and analyzed at the Project Data Centre in Vancouver, BC. All participating cohorts have received approval from their institutional ethics boards to contribute non-nominal patient-specific data to CANOC.

In this analysis we excluded individuals with fewer than 90 days of follow-up to account for immortal person-time since these individuals were never at risk for the outcome of interest (a TI ≥ 90 days). The cohort was administratively censored September 1^st, 2011.

Outcomes

The first outcome, time to first TI, was defined as an interruption of all antiretroviral drugs for a period of at least 90 consecutive days, treated as time-varying into an absorbent state (once interrupted always interrupted). TIs were identified through prescription refill information in BC and through a mixture of clinician reports and pharmacy information for sites in Ontario and Quebec. Treatment interruption is distinct from loss-to-follow-up, which is defined here as no clinical or laboratory contact or prescription refills by participants for 18 months. We performed a sensitivity analysis to assess the impact of differences in TI ascertainment (clinician report vs. pharmacy information) on time to first TI. Time origin for this aim was cART initiation and the time axis used was time since cART initiation. A secondary outcome was time to resumption of cART, defined as the time from initiation of first TI (time origin) to time of treatment resumption for participants who had at least one interruption.

Statistical Analyses

Descriptive analyses were used to study trends of TIs over time and the chi-squared test was used to determine whether there were trends in the frequency of TIs within one year of cART initiation, among individuals with at least 12 months of follow-up from 2000 to 2011. We also examined the proportion of CANOC participants with at least 12 months of follow-up who initiated and interrupted cART in the same calendar year to demonstrate temporal trends. Patient demographical and clinical characteristics at cART initiation were tabulated by treatment interruption status and examined for differences using chi-square statistics for categorical data and Wilcoxon’s rank sum test for continuous variables.

Univariate Cox proportional hazards models were used to examine the relationship of covariates with the outcome. Covariates of interest included age at treatment initiation (per 10 year increase), sex, Aboriginal ancestry, province, injecting drug use (IDU) history, and clinical variables such as composition of initial cART regimen (nucleoside reverse transcriptase inhibitor (NRTI) backbone and third drug in the regimen), hepatitis C positive, and AIDS, HIV plasma viral load (log₁₀) and CD4 cell count at cART initiation. Participants were classified as ‘ever HCV co-infected’ if identified as HCV-positive through physician reports, antibody test results, or PCR test results. Information on pregnancy and treatment in primary infection was not available in the dataset.

Covariates with p-values less than <0.05 and those deemed important based on a priori information were considered candidates for the multivariable model. Multivariable Cox proportional hazards models were constructed to examine factors associated with time to first TI. The time-to-TI analysis was limited to individuals who initiated treatment after January 1^st, 2006, to assess only the effect of unstructured TIs (excluding structured TIs, which ceased in 2006). Cox regression was also used to examine factors associated with time to resumption of cART after a first interruption from 2000–2011, with time origin at first treatment interruption. The proportional hazards assumption was met for both endpoints. Akaike Information Criteria, which balances model goodness of fit with the number of parameters, was used to select the models which best fit the data for all analyses (lower AIC indicates better fit). Analyses were performed using Stata statistical software, version 12.1 [26].

Results

Data were available on 7,633 individuals with at least 90 days of follow-up, of whom 577 (8%) died during 34,921 person-years of follow-up. 1,860 participants (24.5%) interrupted cART over the study period from 2000–2011. Overall, 81% of the sample was male, 22% IDU, 67% MSM, and 45% resided in BC. Figure 1 shows the decrease in TIs in the first calendar year of treatment among CANOC participants over the study period. TIs declined every year from 14% in the year 2000 until 2006, at which point the proportion of new initiators interrupting treatment in their first calendar year of cART leveled off to approximately 7%. Similarly, TIs in the first year of treatment (based on the first 365 days of treatment, data not shown) declined from 11% in 2000 to 8% in 2009. Of all first TIs, 681 (37%) occurred within the first 6 months after cART initiation; 290 (16%) occurred between 6–12 months; 351 (19%) between 1–2 years and 538 (29%) occurred ≥2 years after cART initiation. In terms of duration of TIs, 604 (33%) individuals interrupted for fewer than six months, 418 (23%) interrupted between six months and a year, 285 (15%) of individuals interrupted for one to two years, and 553 (30%) had TIs longer than two years. Of the 1,860 individuals who interrupted, 1,221 (16%) interrupted once, 371 (5%) had two interruptions, 158 (2%) had three interruptions and 110 (6%) had between four and nine interruptions.

Figure 1.

Proportion of CANOC participants with at least 12 months of follow-up who interrupted combination antiretroviral therapy (cART) for at least 3 months within one year of initiation, by calendar year (N=6,463)

Table 1 shows patient demographic and clinical characteristics at cART initiation by treatment interruption status in the CANOC collaboration. Compared to non-interrupters, treatment interrupters were more likely to be female (31% vs. 15%), younger (median age, interquartile range (IQR): 38 (32–44) vs. 41 (34–47)), report Aboriginal ancestry (13% vs. 3%), and live in BC (64% vs. 39%) (all p<0.001). Behavioral and clinical factors more commonly reported by interrupters were a history of IDU (43% vs. 16%), heterosexual transmission (33% vs. 25%), hepatitis C seropositivity (46% vs. 18%), use of zidovudine/lamivudine (33% vs. 21%) and stavudine/lamivudine (16% vs. 7%) as initial NRTI combinations and use of nevirapine (20 vs. 9%) and an “other” third drug in the cART regimen at initiation (all p<0.001). Interrupters were less likely to report AIDS prior to cART (13% vs. 14%, p=0.005), use abacavir/lamivudine (10% vs. 16%, p<0.001) or tenofovir/emtricitabine (20% vs. 41%, p<0.001) and efavirenz (25% vs. 38%, p<0.001), or atazanivir (16% vs. 23%) as NRTI combinations and third drugs in the initial cART regimen, respectively.

Table 1.

Baseline characteristics of 7,633 CANOC participants from 2000–2011 by treatment interruption status

Characteristic	N	No treatment Interruption n (%) n = 5,773	Treatment Interruption n (%) n = 1,860	p-value
Sex	7,633
Male		4,904 (85.0)	1,287 (69.2)	<0.001
Female		869 (15.1)	573 (30.8)
Median age at baseline (IQR)*	7,633	41 (34–47)	38 (32–44)	<0.001
Province	7,633
British Columbia		2,276 (39.4)	1,183 (63.6)	<0.001
Ontario		2,135 (37.0)	422 (22.7)
Quebec		1,362 (23.6)	255 (13.7)
Aboriginal ancestry	7,633
Yes		172 (3.0)	242 (13.0)	<0.001
No		836 (14.5)	441 (23.7)
Unknown		4,765 (82.5)	1,177 (63.3)
History of injecting drug use	7,633
Yes		906 (15.7)	790 (42.5)	<0.001
No		3,477 (60.2)	780 (41.9)
Unknown		1,390 (24.1)	290 (15.6)
AIDS prior to cART‡	7,633
Yes		791 (13.7)	237 (12.7)	0.005
No		4,521 (78.3)	1,513 (81.3)
Unknown		461 (8.0)	110 (5.9)
Hepatitis C	7,633
Yes		1,024 (17.7)	851 (45.8)	<0.001
No		4,338 (75.1)	902 (48.5)
Unknown		411 (7.1)	107 (5.8)
Median CD4 cell count (IQR)	7,633	210 (120–300)	210 (110–310)	0.004
Median pVL**(log10) (IQR)	7,633	4.87 (4.33–5.00)	4.84 (4.29–5.00)	<0.001
Year initiated cART	7,633
2000–2003		1,309 (22.7)	878 (47.2)	<0.001
2004–2006		1,520 (26.3)	526 (28.3)
2007–2011		2,944 (51.0)	456 (24.5)
NRTI^ combo in baseline
cART‡ regimen	7,633			<0.001
Tenofovir/Emtricitabine		2,375 (41.1)	363 (19.5)
Zidovudine/Lamivudine		1,223 (21.2)	619 (33.3)
Tenofovir/Lamivudine		458 (7.9)	157 (8.4)
Abacavir/Lamivudine		905 (15.7)	177 (9.5)
Stavudine/Lamivudine		395 (6.8)	292 (15.7)
Other		417 (7.2)	252 (13.6)
Third drug in baseline
cART‡ regimen	7,633			<0.001
Nevirapine		507 (8.8)	371 (20.0)
Efavirenz		2,193 (38.0)	470 (25.3)
Lopinavir		1,118 (19.4)	351 (18.9)
Atazanavir		1,347 (23.3)	296 (15.9)
Other		608 (10.5)	372 (20.0)

^*IQR= interquartile range

^**pVL= HIV plasma viral load

^{^}NRTI= nucleoside reverse transcriptase inhibitor

^‡cART= combination antiretroviral therapy

Time to first TI

The Cox regression model evaluating factors predicting time to first TI was restricted to individuals initiating cART from 2006 onwards (to exclude all instances of structured TIs) and is presented in Table 2. Data were available for 4,134 individuals, of whom 626 (15%) had at least one TI over 9,833 person-years. Mean follow up in this analysis was 2.4 years and median time to first TI was 0.60 years (7.2 months). Predictors of a first TI were female sex (adjusted hazard ratio (aHR): 1.59, 95% CI: 1.33–1.92), Aboriginal ancestry (aHR: 1.67, CI: 1.27–2.20), a history of injecting drug use (aHR: 1.43, CI: 1.09–1.89), hepatitis C seropositivity (aHR: 2.17, CI: 1.68–2.79), a baseline CD4 cell count above 350 cells/mm³ versus less than 200 cells/mm³ (aHR: 1.46, CI: 1.17–1.81) and use of zidovudine versus tenofovir in the initial cART regimen (aHR: 2.47, CI: 1.92–3.20). Factors protective against TI were older age (aHR: 0.79 per 10 year increase, CI: 0.73–0.87), higher HIV plasma viral load (log₁₀) (aHR: 0.87, CI: 0.78–0.97) and residence in Ontario (aHR: 0.55, CI: 0.43–0.70) or Quebec (aHR: 0.42, CI: 0.31–0.57) versus BC. There was some collinearity demonstrated in the model between IDU and hepatitis C variables as well as plasma viral load and CD4 cell count that attenuated both HRs. Figure 2 shows that the cumulative probability of interrupting treatment over the study period was 0.10 (CI: 0.09–0.11) after one year on cART.

Table 2.

Factors predicting time to first TI among CANOC participants who initiated treatment from 2006 to 2011 (N=4,134)

	Unadjusted hazard ratio (95% confidence interval)	p-value	Adjusted hazard ratio (95% confidence interval)	p-value
Female vs. male	2.78 (2.34–3.13)	<0.001	1.59 (1.33–1.92)	<0.001
Age (per 10 year increment)	0.78 (0.71–0.85)	<0.001	0.79 (0.73–0.87)	<0.001
Province
British Columbia	1.00		1.00
Ontario	0.46 (0.38–0.57)	<0.001	0.55 (0.43–0.70)	<0.001
Quebec	0.29 (0.22–0.39)	<0.001	0.42 (0.31–0.57)	<0.001
Aboriginal ancestry
No	1.00		1.00
Yes	3.12 (2.37–4.11)	<0.001	1.69 (1.27–2.20)	<0.001
Unknown	0.59 (0.47–0.73)	<0.001	1.01 (0.90–1.28)	0.932
History of injecting drug use
No	1.00		1.00
Yes	3.92 (3.27–4.70)	<0.001	1.43 (1.08–1.89)	0.010
Unknown	1.14 (0.90–1.45)	0.285	1.02 (0.79–1.31)	0.874
AIDS prior to cART*
No	1.00
Yes	0.93 (0.71–1.21)	0.597
Unknown	0.60 (0.39–0.93)	0.023
Hepatitis C
No	1.00
Yes	3.81 (3.22–4.51)	<0.001	2.17 (1.68–2.79)	<0.001
Unknown	0.99 (0.66–1.48)	0.958	0.75 (0.50–1.12)	0.160
CD4 cell count at treatment initiation
<200 cells/mm3	1.00		1.00
200–349 cells/mm3	0.76 (0.63–0.92)	0.004	0.96 (0.80–1.16)	0.688
350+ cells/mm3	1.27 (1.02–1.57)	0.030	1.46 (1.17–1.81)	<0.001
Plasma viral load at initiation (log10)	0.84 (0.77–0.93)	0.001	0.87 (0.78–0.97)	0.015
Year initiated cART*
2006	1.00
2007	0.84 (0.68–1.04)	0.110
2008	0.67 (0.53–0.85)	0.001
2009	0.67 (0.52–0.87)	0.002
NRTI** combo in baseline cART* regimen
Tenofovir/Emtricitabine	1.00		1.00
Zidovudine/Lamivudine	2.41 (1.89–3.08)	<0.001	2.47 (1.92–3.20)	<0.001
Tenofovir/Lamivudine	1.26 (0.95–1.67)	0.110	1.32 (1.00–1.75)	0.052
Abacavir/Lamivudine	0.85 (0.68–1.08)	0.189	1.22 (0.97–1.53)	0.089
Stavudine/Lamivudine	1.17 (0.48–2.83)	0.729	1.63 (0.67–3.99)	0.282
Other	1.20 (0.72–2.02)	0.481	1.64 (0.97–2.80)	0.066
Third drug in baseline cART* regimen
Nevirapine	1.00
Efavirenz	1.63 (0.97–2.71)	0.061
Lopinavir	2.08 (1.23–3.50)	0.006
Atazanavir	1.72 (1.03–2.87)	0.037
Other	3.47 (1.93–6.22)	<0.001

^*cART= combination antiretroviral therapy

^**NRTI= nucleoside reverse transcriptase inhibitor

Figure 2.

Cumulative proportion of CANOC participants interrupting treatment after cART initiation (2006–2011)

Time to cART resumption

Of 1,860 individuals who interrupted therapy, 1,566 (84%) eventually restarted cART. Median time to cART resumption was 9.6 months. The Cox regression model examining factors predicting resumption of treatment after a first TI, presented in Table 3, shows that male sex (aHR: 1.22, CI: 1.09–1.37), older age (aHR: 1.10 per 10 year increase, CI: 1.04–1.17) and initiation of cART in 2004–2006 (aHR: 1.27, CI: 1.11–1.45) or 2007–2011 (aHR: 1.45, CI: 1.24–1.70) versus 2000–2003 were significantly associated with treatment resumption. A CD4 cell count 200–349 cells/mm³ (aHR: 0.76, CI: 0.68–0.85) and greater than 350 cells/mm³ (aHR: 0.43, CI: 0.37–0.50) at cART initiation compared to a CD4 cell count less than 200 cells/mm³ and residence in Ontario (aHR: 0.81, CI: 0.71–0.92) versus BC were associated with less likely resumption of treatment.

Table 3.

Factors predicting time to resumption of cART after a first TI among 1,860 CANOC participants who initiated and interrupted treatment from 2001–2011

	Unadjusted hazard ratio (95% confidence interval)	p-value	Adjusted hazard ratio (95% confidence interval)	p-value
Female vs. male	0.72 (0.65–0.81)	<0.001	0.82 (0.73–0.92)	0.001
Age (per 10 year increment)	1.19 (1.13–1.26)	<0.001	1.10 (1.04–1.17)	0.001
Province
British Columbia	1.00		1.00
Ontario	0.70 (0.62–0.80)	<0.001	0.81 (0.71–0.92)	0.002
Quebec	0.77 (0.66–0.89)	0.001	0.88 (0.75–1.03)	0.112
Aboriginal ancestry
No	1.00
Yes	0.92 (0.78–1.09)	0.350
Unknown	0.78 (0.69–0.87)	<0.001
History of injecting drug use
No	1.00
Yes	1.19 (1.07–1.33)	0.001
Unknown	0.96 (0.83–1.12)	0.634
AIDS prior to cART*
No	1.00
Yes	1.45 (1.25–1.67)	<0.001
Unknown	0.87 (0.70–1.08)	0.218
Hepatitis C
No	1.00
Yes	1.12 (1.01–1.24)	0.025
Unknown	0.76 (0.60–0.97)	0.028
CD4 cell count at treatment initiation
<200 cells/mm3	1.00		1.00
200–349 cells/mm3	0.75 (0.67–0.84)	<0.001	0.76 (0.68–0.85)	<0.001
350+ cells/mm3	0.40 (0.35–0.46)	<0.001	0.43 (0.37–0.50)	<0.001
Year initiated cART*
2000–2003	1.00		1.00
2004–2006	1.25 (1.11–1.40)	<0.001	1.27 (1.11–1.45)	0.001
2007–2011	1.42 (1.25–1.62)	<0.001	1.45 (1.24–1.70)	<0.001
NRTI** combo in baseline cART* regimen
Tenofovir/Emtricitabine	1.00
Zidovudine/Lamivudine	0.55 (0.47–0.64)	<0.001
Tenofovir/Lamivudine	0.93 (0.76–1.13)	0.457
Abacavir/Lamivudine	0.89 (0.73–1.10)	0.267
Stavudine/Lamivudine	0.74 (0.62–0.88)	<0.001
Other	0.59 (0.49–0.70)	<0.001
Third drug in baseline cART* regimen
Nevirapine	1.00		1.00
Efavirenz	1.22 (1.05–1.41)	0.011	0.96 (0.81–1.14)	0.640
Lopinavir	1.27 (1.08–1.49)	0.003	1.16 (0.97–1.38)	0.095
Atazanavir	1.59 (1.35–1.88)	<0.001	1.02 (0.84–1.25)	0.838
Other	0.83 (0.71–0.97)	0.017	0.88 (0.75–1.04)	0.124

^*cART= combination antiretroviral therapy

^**NRTI= nucleoside reverse transcriptase inhibitor

Discussion

Our results demonstrate that the frequency of TI remains relatively high in a setting of universal free access to HIV care. TIs continue to be pervasive, with 25% of CANOC participants reporting at least one interruption of at least 90 days over the study period. However, it is reassuring that the proportion of individuals interrupting cART within the first calendar year of treatment decreased over time and stabilized over the last five years. Collectively, these data suggest that there is an urgent need to strengthen communication about the need to embrace sustained, lifelong therapy with health providers and patients, as a means to optimize the cART benefits of reduced morbidity, mortality and transmission.

Improvements in ease of delivery and the side effect profile of cART over time appear to have improved the acceptability of these medications for patients. Notably, we have documented a marked decline in the proportion of individuals interrupting treatment in their first year of treatment over the study period. These results are complementary to data from BC which reported a reduction in the prevalence of TIs over the period from 2000–2006 [5]. While medication side effects were the primary reason for TIs at one time [27–30], decreased toxicity associated with cART component drugs has improved the tolerability of these medications considerably [30–33]. Consistent with this information, we demonstrated that patients who initiate on certain older drugs such as zidovudine have a higher risk of interrupting treatment. Additionally, the preponderant use of compact once daily fixed dose formulations as preferred first line regimens may have contributed to the observed reduction in the proportion of TIs in the first year of treatment in more recent years [34], as are 2006 guidelines discontinuing physician-directed TIs.

Nonetheless, it is concerning that our analysis highlighted a higher risk of TI among individuals who initiate treatment earlier in the course of their disease, demonstrated by higher CD4 cell counts and lower plasma viral load pre-cART, as has been observed in previous studies [16, 24, 35–38]. Additionally, we found that individuals with higher CD4 cell counts are less likely to reinitiate therapy after interruption. In this context, it is important to emphasize that the current analyses were undertaken using data from 2000–2011, during which time cART guidelines were evolving. It is only now, in 2013, that consensus has been reached regarding the use of cART at CD4 counts ≤500 cells/mm³ and in selected additional populations regardless of CD4 count (such as serodiscordant couples, pregnant women and individuals with co-existing tuberculosis) [19–23]. Therefore, it will be important to closely prospectively monitor the incidence and determinants of TIs for the foreseeable future, as the number of HIV-positive individuals initiating treatment with higher cell counts and lower viral load is likely to increase. Assuring that these relatively less immunosuppressed patients are provided with the support to maintain continuity of treatment will be imperative to avoid the development of drug resistance and worsened prognosis due to TIs as the HIV “treatment as prevention” strategy is more widely implemented [31]. Moreover, as many of these newly eligible individuals will continue treatment for an extended period of time, strategies to counter treatment fatigue are urgently needed.

Our results suggest that women are at increased risk of interrupting treatment and less likely to resume cART following TI. However, this has not been a consistent finding across studies [5, 24, 39–41]. Higher risk in women is a phenomenon that may partly be due to initiation of cART during pregnancy and subsequent discontinuation of treatment. Other possible explanations are that women may report more cART side-effects and toxicities [40, 42], that they may have greater drug intolerance during pregnancy [43] and that women may be unwilling to engage consistently with or prioritize cART if doing so may lead to negative perceptions or neglect from others [44].

Consistent with previous studies, we found that younger individuals [5, 35, 45, 46], those initiating cART with a regimen containing zidovudine [5], and those with hepatitis C co-infection [5, 24] and a history of IDU [5, 24, 35] were more likely to interrupt cART. Regional differences were also observed, with less risk of TI in Ontario and Quebec compared to BC, a finding likely explained by inherent cohort differences. For instance, BC data represents all HIV-positive individuals on cART across the province while Ontario and Quebec data are derived from a selection of specialized clinics; in addition, there are differences in the proportion of women, Aboriginal individuals and IDU. Lastly, individuals self-reported as having Aboriginal ancestry were at higher risk of interrupting treatment. Aboriginal peoples are over-represented in the HIV epidemic in Canada, constituting 4% of the population according to the 2011 Canadian census and 12% of new HIV infections in 2011 [47, 48]. Additional knowledge should be generated to understand the reasons for interruption for each of these specific sub-populations and specialized services should be designed to support continuous engagement of treatment accordingly.

Resumption of treatment, observed in more than three quarters of individuals who interrupted, was more likely in individuals who were males, resided in BC, were older, initiated cART more recently and had a CD4 cell count <200 cells/mm³ at treatment initiation. Thus, certain predictors of TI were the opposite of predictors of resumption, with women and those with higher CD4 cell counts more likely to interrupt and men and those with lower CD4 cell counts more likely to resume cART. Other studies have found similar rates of resumption [5, 41, 49]. While the probability of restarting cART is relatively high, it is important to note that 15% of individuals never reinitiated treatment while under observation. This remains a substantial concern, as these individuals are at increased risk of disease progression and premature death, as well as at increased risk of transmitting HIV infection. As with risk of TI, younger, female individuals initiating cART with higher CD4 cell counts should be targeted for additional support to increase the likelihood of resumption after TI.

Retention in treatment represents a crucial indicator of long-term success for people living with HIV. Risk factors for TI identified in this study will aid in identifying populations in need of increased support. A number of useful guidelines for managing patient retention in care have been developed specifically addressing adherence [50] and may also be useful in the context of TIs. The findings presented here may be useful in the development of interventions specific to interruptions in treatment; this is an opportune area for future research. In particular, more than half of first TIs occurred in the first year of treatment, which suggests that new initiators are particularly vulnerable to prolonged gaps in engagement and could be targeted for increased monitoring and support.

Strengths and limitations

Using data from the largest collaboration of HIV cohorts in Canada to inform analyses lent considerable power to analyses. CANOC represents about a quarter of all HIV-positive individuals in the country who are currently on cART and approximately half of those who have initiated on more modern regimens since 2000 [51]. This improves the ability to generalize study results to the remainder of the HIV-positive population in Canada currently accessing treatment. Unlike many other studies examining interruptions in treatment, this study focused on the modern cART period and restricted analyses pertaining to predictors of TIs to the period following the recommendation not to prescribe TIs, from 2006 onwards. CANOC also enrolled cART-naïve individuals, which helps isolate the effect of the latest treatment regimens on risk of TI.

This study has several limitations. Ascertainment of TI differs among CANOC sites, with some sites recording TIs from medical charts (data from physician, nurses and pharmacists), and others using cART prescription refill information as well as information from medical charts to identify TIs. We sought to mitigate this potential bias by examining differences in time to first TI comparing sites using medical charts versus those that used pharmacy and medical charts. No differences were observed (data not shown); however, residual confounding may be present that may also help explain regional differences in risk of TI. There may also be some degree of underreporting of TIs in Quebec in Ontario. We were not able to measure some factors which have been associated in other studies with TIs, such as pregnancy, co-morbidities, depression [45, 52], mental disorders [37], incarceration [53–55], treatment in primary infection and socioeconomic factors such as unstable housing and lack of social supports; these were not captured in the CANOC database. As data become available, it will be interesting to observe the impact of these latter factors. Future research studying TIs should seek to examine these factors and their contribution to the incidence and prevalence of TIs. Lastly, CANOC includes data from only three provinces, and a clinic-based selection bias exists, as included data from BC includes the entire sample of people on cART province-wide while data from Ontario and Quebec come from a selection of clinics that are mainly HIV-specific.

In conclusion, our results demonstrate that TIs remain relatively prevalent. Specific strategies may be warranted to decrease the incidence of TIs, particularly targeting the most affected groups, such as those who are women, younger, of Aboriginal ancestry, IDU, less immunosuppressed at treatment initiation, initiate cART with zidovudine/lamivudine, HCV co-infected and have lower HIV plasma viral load at cART initiation. Our results uncovered the use of zidovudine versus tenofovir in the initial cART regimen as a potentially amenable factor associated with TIs. Full implementation of current guidelines, which clearly favor the use of tenofovir over zidovudine as a first line agent for cART, should help address this issue. On the other hand, the finding that earlier initiation of cART was associated with higher rates of TIs is concerning, and merits close prospective monitoring as new guidelines are widely and fully implemented. Finally, while it is reassuring to see that over three quarters of the observed TIs were time-limited, in view of the serious individual and societal consequences of TIs, further targeted efforts are required to maximize resumption of sustained cART as soon as possible following TIs.

Appendix

*The CANOC Collaboration includes: Investigators: Gloria Aykroyd (Ontario HIV Treatment Network, OHTN), Louise Balfour (University of Ottawa, OHTN Cohort Study, OCS Co-Investigator), Ahmed Bayoumi (University of Toronto, OCS Co-Investigator), Ann Burchell (Ontario HIV Treatment Network), John Cairney (University of Toronto, OCS Co-Investigator), Liviana Calzavara (University of Toronto, OCS Co-Investigator), Angela Cescon (British Columbia Centre for Excellence in HIV/AIDS), Curtis Cooper (University of Ottawa, OCS Co-Investigator), Kevin Gough (University of Toronto, OCS Co-Investigator), Silvia Guillemi (British Columbia Centre for Excellence in HIV/AIDS, University of British Columbia), P. Richard Harrigan (British Columbia Centre for Excellence in HIV/AIDS, University of British Columbia), Marianne Harris (British Columbia Centre for Excellence in HIV/AIDS), George Hatzakis (McGill University), Robert Hogg (British Columbia Centre for Excellence in HIV/AIDS, Simon Fraser University), Sean Hosein (CATIE), Don Kilby (University of Ottawa, Ontario HIV Treatment Network), Marina Klein (Montreal Chest Institute Immunodeficiency Service Cohort, McGill University), Richard Lalonde (The Montreal Chest Institute Immunodeficiency Service Cohort and McGill University), Viviane Lima (British Columbia Centre for Excellence in HIV/AIDS, University of British Columbia), Mona Loutfy (University of Toronto, Maple Leaf Medical Clinic, OCS Co-Investigator), Nima Machouf (Clinique Medicale l’Actuel, Université de Montréal), Ed Mills (British Columbia Centre for Excellence in HIV/AIDS, University of Ottawa), Peggy Millson (University of Toronto, OCS Co-Investigator), Julio Montaner (British Columbia Centre for Excellence in HIV/AIDS, University of British Columbia), David Moore (British Columbia Centre for Excellence in HIV/AIDS, University of British Columbia), Alexis Palmer (British Columbia Centre for Excellence in HIV/AIDS), Janet Raboud (University of Toronto, University Health Network, OCS Co-investigator), Anita Rachlis (University of Toronto, OCS Co-Investigator), Stanley Read (University of Toronto, OCS Co-Investigator), Sean Rourke (Ontario HIV Treatment Network, University of Toronto), Hasina Samji (British Columbia Centre for Excellence in HIV/AIDS), Marek Smieja (McMaster University, OCS Co-Investigator), Irving Salit (University of Toronto, OCS Co-Investigator), Darien Taylor (Canadian AIDS Treatment Information Exchange, OCS Co-Investigator), Benoit Trottier (Clinique Medicale l’Actuel, Université de Montréal), Chris Tsoukas (McGill University), Sharon Walmsley (University of Toronto, OCS Co-Investigator), and Wendy Wobeser (Queens University, OCS Co-Investigator). Analysts and Staff: Mark Fisher (OHTN), Sandra Gardner (University of Toronto), Nada Gataric (British Columbia Centre for Excellence in HIV/AIDS), Guillaume Colley (British Columbia Centre for Excellence in HIV/AIDS), Sergio Rueda (OHTN), and Benita Yip (British Columbia Centre for Excellence in HIV/AIDS).