MAJOR ARTICLE

Time-Dependent Predictors of Loss to Follow-Up in a Large HIV Treatment Cohort in Nigeria
Seema Thakore Meloni,1 Charlotte Chang,1 Beth Chaplin,1 Holly Rawizza,1,2 Oluwatoyin Jolayemi,3 Bolanle Banigbe,3 Prosper Okonkwo,3 and Phyllis Kanki1
1

Harvard School of Public Health, and 2Brigham and Women’s Hospital, Boston, Massachusetts; and 3AIDS Prevention Initiative Nigeria, Ltd./Gte., Abuja

Background. Most evaluations of loss to follow-up (LTFU) in human immunodeficiency virus (HIV) treatment programs focus on baseline predictors, prior to antiretroviral therapy (ART) initiation. As risk of LTFU is a continuous issue, the aim of this evaluation was to augment existing information with further examination of time-dependent predictors of loss. Methods. This was a retrospective evaluation of data collected between 2004 and 2012 by the Harvard School of Public Health and the AIDS Prevention Initiative in Nigeria as part of PEPFAR-funded program in Nigeria. We used multivariate modeling methods to examine associations between CD4+ cell counts, viral load, and early adherence patterns with LTFU, defined as no refills collected for at least 2 months since the last scheduled appointment. Results. Of 51 953 patients initiated on ART between 2004 and 2011, 14 626 (28%) were LTFU by 2012. Factors associated with increased risk for LTFU were young age, having nonincome-generating occupations or no education, being unmarried, World Health Organization (WHO) stage, having a detectable viral load, and lower CD4+ cell counts. In a subset analysis, adherence patterns during the first 3 months of ART were associated with risk of LTFU by month 12. Conclusions. In settings with limited resources, early adherence patterns, as well as CD4+ cell counts and unsuppressed viral load, at any time point in treatment are predictive of loss and serve as effective markers for developing targeted interventions to reduce rates of attrition. Keywords. HIV; ART outcomes; loss to follow-up; retention; attrition.

The successful global scale-up of antiretroviral therapy (ART) programs has vastly changed outcomes for patients infected with human immunodeficiency virus (HIV) throughout sub-Saharan Africa (SSA), shifting the disease from one that was once uniformly fatal to a chronic disease with lifelong ART. An integral factor in redefining the infection from an acute to manageable condition is the commitment from patients that once they are initiated on an ART regimen, they will remain

Received 4 June 2014; accepted 20 June 2014. Correspondence: Phyllis Kanki, DVM, DSc, Harvard School of Public Health, Department of Immunology and Infectious Diseases, 651 Huntington Ave, FXB 405, Boston, MA 02115 (pkanki@hsph.harvard.edu). Open Forum Infectious Diseases © The Author 2014. Published by Oxford University Press on behalf of the Infectious Diseases Societyof America. This is an Open Access article distributed under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs licence (http:// creativecommons.org/licenses/by-nc-nd/4.0/), which permits non-commercial reproduction and distribution of the work, in any medium, provided the original work is not altered or transformed in any way, and that the work is properly cited. For commercial re-use, please contact journals.permissions@oup.com. DOI: 10.1093/ofid/ofu055

adherent. Once a patient initiates a treatment regimen, they must return for clinical visits, laboratory tests, prescription refills, and counseling services. Oftentimes, these requirements become burdensome, and patients discontinue services, creating a major challenge for most HIV treatment programs. With approximately 168 million inhabitants, Nigeria is the most populous country in SSA and has maintained an HIV prevalence of approximately 4% for the past 6 years [1–3]. The government of Nigeria initiated its National ART Program in 2001 and later gained support from the President’s Emergency Plan For AIDS Relief (PEPFAR) and other international donor agencies in mid-2000s. From 2004 to 2012, the Harvard School of Public Health partnered with the AIDS Prevention Initiative in Nigeria (APIN) to use PEPFAR funding provided through the US Health Resources and Service Administration (HRSA) to scale up HIV prevention, care, and treatment activities across Nigeria. The scale-up of HIV care has been a major success. However, the need to initiate over a million additional
Time-Dependent Predictors of LTFU

•

OFID

•

1

patients on ART, while retaining those already on therapy despite tightening budgets, represents a significant challenge; thus, cost-effective markers for developing and monitoring targeted interventions are needed. Rates and predictors of LTFU have been evaluated in numerous HIV programs, and rates range from 0.3% to 50% [4–18]. Although the studies differ on identified predictors of LTFU, the majority focus on demographic and baseline measures taken prior to ART initiation and do not evaluate predictors in a time-dependent fashion [17,18]. The goal of this evaluation was to use data from the Harvard/APIN PEPFAR program, along with statistical methods that incorporate time-dependent factors, thus building upon existing information on baseline predictors of LTFU already in the literature; the hypothesis was that adherence patterns, CD4+ cell counts and viral loads (VLs) would be strong indicators of risk of LTFU in a timedependent manner. The ultimate goal in addressing this hypothesis was to identify an easy and inexpensive method for healthcare works challenged with limited resources and high patient volumes to identify candidates for targeted interventions to improve retention. METHODS
Patients

or transferred to non-Harvard/APIN sites during the period of evaluation were not considered LTFU.
Factors Associated With LTFU

We evaluated baseline demographic (age, sex, education, occupation type, enrollment site, enrollment year, and HIV transmission category) and clinical (hepatitis B virus [HBV] and/ or hepatitis C virus [HCV] coinfection at enrollment, WHO clinical stage, ART regimen, CD4+ cell count, and VL) factors, where baseline is defined as at the time of ART initiation. For analyses, age was converted to a categorical variable based on quartiles, and the occupation category was collapsed into non-income-generating (ie, unemployed, students, job applicants, housewives/homemakers, and retirees) and incomegenerating (laborer vs professional) categories. Additionally, we incorporated time-dependent factors into the analyses, including adherence patterns during the first 3 months of treatment, CD4+ cell count, and VL.
Measurement of Adherence

Upon entry into the Harvard/APIN PEPFAR HIV care program and following informed consent, all patients were assessed for ART eligibility according to Nigerian National Guidelines, which followed the WHO guidelines [19,20]. All consent forms were approved by the institutional review boards at Harvard, APIN, and all the corresponding Harvard/APIN PEPFAR HIV care and treatment sites. All ART-eligible patients were placed on ART following a clinical examination and a set of baseline laboratory tests, which included hematology, clinical chemistries, CD4+ cell count, and VL enumeration. Patients were generally given a 30-day supply of antiretroviral (ARV) medications. Following the first prescription pick-up, refills were obtained on a monthly basis. Following the initiation of ART, laboratory tests were repeated every 6 months unless an earlier evaluation was medically necessary. All patient data were maintained in electronic databases. For the analyses, we included patients who were enrolled on ART between June 2004 and February 2011 to ensure at least 1 year of follow-up time for the evaluation. All patients were at least 15 years of age at enrollment. Patients who had previous ARV experience prior to enrolling in the Harvard/APIN program were excluded.
Definition of Loss to Follow-Up

To evaluate the time-specific association between adherence and LTFU, we focused on adherence patterns during the first 3 months of ART to determine whether the pattern was predictive of LTFU by month 12. For these subset evaluations, we excluded patients that discontinued during the first 3 months of treatment in order to avoid biasing values with those who had poor adherence during early treatment. We used prescription refill timeliness as the measurement of adherence, which has been previously shown in multiple studies to be a strong surrogate [21–26]. To compute pill coverage for the first 90 days of treatment, we divided the total number of pills supplied for the time period by 90 and then multiplied by 100 for an average percent adherence during the time period. Average percent adherence values were collapsed into categories for analyses.
Statistical Analyses

Patients were classified as LTFU if, at the time of interest, at least 2 months had elapsed since the patient’s last scheduled pick-up date and they did not later return. Patients who died, withdrew,
2

Univariate comparisons categorical variables were performed using the χ2 or Fisher exact test; Student t-test was used for normally distributed continuous variables and the Wilcoxon rank sum test for non-normally distributed continuous variables. Statistical significance was defined at an α-level of 0.05. Categorical variables were collapsed based on results of univariate analyses. Kaplan-Meier analyses were used to examine the probability of follow-up for patients that initiated ART between June 2004 and February 2011. Patients were considered at risk of LTFU from the time they initiated ART to the date of their last pick-up, transfer, withdrawal, or death. Patients who withdrew or transferred were censored at the date of their last pick-up, and patients who died were censored at the time of their death. Cox proportional hazards models were used to evaluate

•

OFID

•

Meloni et al

baseline and demographic factors associated with LTFU. Unmeasured heterogeneity between sites was controlled in the models by using random effects methods. Additionally, clinically relevant interaction terms were tested as potentially explanatory of significant findings. To address potential bias due to patients who were excluded because of missing data, multiple imputation of missing values were performed using chained equations assuming missing at random and 10 imputed data sets. To further evaluate the time-dependent association between CD4+ cell counts and risk of LTFU, we examined LTFU in yearly increments following ART initiation, starting with the second pick-up for the CD4+ cell count. For the first time point (ie, “after visit 1”), we compared median baseline CD4+ cell counts of those who were lost following the first visit to those who were retained beyond the first visit. For each subsequent time point (ie, months 12, 24, 36, and 48), we compared median CD4+ cell counts from the visit 6 months prior to determine if it was predictive of LTFU or retention by the noted time point. To examine the time-dependent association between VL and LTFU, we similarly analyzed retention at months 12, 24, 36, and 48 and compared VL suppression rates at the visit 6 months prior to determine if suppression was predictive of LTFU. For both the CD4+ cell count and VL evaluations, each time point contained data from those patients retained in the prior time point; patients who transferred, withdrew, or died in the prior time period were removed from subsequent cohorts. To determine whether the relationships between CD4+ cell counts and viral loads with LTFU rates remained after adjusting for other predictors of LTFU, we generated random effects Cox proportional hazard models including CD4+ cell counts and viral load suppression as time-varying covariates, while controlling for all other significant baseline and demographic predictors of LTFU. Values for the Cox models were generated using both complete cases and multiply imputed data. Multiple imputations for time-dependent data were generated using a 2-fold fully conditional specification algorithm for imputation of missing longitudinal data. For the subanalysis on the association between adherence patterns during the first 3 months of treatment and subsequent risk of LTFU by month 12, we generated a random effects logistic regression model to examine predictors of loss, controlling for site variability. Significant predictors of LTFU from the KaplanMeier analyses were retained in the model regardless of statistical significance as they were shown to be significant predictors of LTFU when all patients were evaluated. All statistical analyses were conducted using Stata version 13 (College Station, TX). RESULTS Between June 2004 and February 2012, 88 983 adult patients initiated standard first-line (1L) ART at one of 32 hospitals

(10 tertiary and 22 secondary) spread across 9 states supported by the Harvard/APIN PEPAR program. Of those patients, 88 665 (99%) were HIV-1 monoinfected. In order to concentrate our analyses on patients with at least 1 year since ART initiation, we focused on the 72 770 patients enrolled as of February 2011. Of those patients, we excluded 15 394 who were ARV-experienced at enrollment (Fig. 1). Of the total 57 376 ARV-naive patients enrolled by February 2011, 4980 (8.7%) were LTFU, 350 died (0.6%), and 93 (0.2%) transferred or withdrew following the first drug pick-up. After comparing patients who did not return after filling their first prescription to those who returned for at least 1 refill and finding these groups to be very different, we also excluded patients who did not return after filling their first prescription (Table 1). Of the 51 953 patients enrolled by February 2011, 14 626 (28%) were LTFU, 816 (2%) died, and 1515 (3%) were reported as transferred or withdrawn following the second drug pick-up as of March 2012. The majority of the lost patients generally discontinued within the first 12–18 months following initiation of ART. When combining data across enrollment years, the retention rates were 91% following the first pick-up, 79% by month 6, 74% by month 12, and 70% by month 18. In evaluating the total percentage lost by time on treatment and year of ART initiation, these rates varied by enrollment year, where loss by month 12 appeared greater for the cohorts enrolled after 2006 as compared to those enrolled between 2004 and 2006. Overall, of the ARV-naive patients with at least 1 prescription refill, 65% were female, 57% had a secondary or tertiary level education, 58% were married, and 75% had income-generating occupations. The median age for the cohort was 35 years (interquartile range [IQR]: 29-41). At baseline, the majority (67%) of patients had a CD4+ cell count of >100 cells/mL. In addition, 78% of patients had a baseline viral load of >10 000 copies/ mL, and 22% had tuberculosis coinfection, 16% were HBsAg positive, and 6% HCV antibody positive.
Baseline Predictors of LTFU

In preliminary adjusted random effects Cox proportional hazard modeling of baseline predictors, controlling for site differences, the factors associated with increased risk for LTFU in the 51 953 patients that made at least 1 refill pick-up were: lower age; being male; initiating ART during or after 2006; having non-income-generating occupations or no education; being single, divorced, or separated; higher baseline WHO clinical stage and viral load; and lower baseline CD4+ cell counts (data not shown). Additionally, we found that patients who started on tenofovir (TDF) + emtricitabine (FTC) or lamivudine (3TC) + efavirenz (EFV) were more likely to be LTFU than those on the zidovudine (AZT)-containing regimens (<0.001). We tested interactions between sex and regimen as well as ART initiation year and regimen and found that neither was significant in explaining model outcomes.
Time-Dependent Predictors of LTFU

•

OFID

•

3

Fig. 1. Flow diagram for patients included in this evaluation. Abbreviations: ART, antiretroviral therapy; ARV, antiretroviral; HIV, human immunodeficiency virus; 1L, first-line. CD4+ Cell Counts and Viral Suppression Rates Predict LTFU in Time-Dependent Manner

After finding that baseline CD4+ cell count was a significant predictor of LTFU, we wanted to determine whether CD4+ cell count remained a continuous predictor of loss. The median CD4+ cell counts (Fig. 2a) continually increased over the 4 years for both of the patients who were retained, as well as those who were not retained in the evaluated time period. For patients that were lost, the median of the last CD4+ cell count prior to loss, regardless of time of loss, was 183 cells/mL (IQR: 86–316 cells/ mL). At each of the assessed time points, the median CD4+ cell count from the prior 6 months was higher in those subsequently retained compared to those LTFU (Fig. 2a). Similarly, we found that retained patients were more likely to be virally suppressed at their prior 6-month visit compared to those who were LTFU (Fig. 2b; P < 0.05); the median VL prior to loss in LTFU patients was 15 457 cp/mL (IQR: 200–143 386), where nearly 71% of patients had detectable viral load within the 6 months preceding loss regardless of time of loss. When we adjusted for age, sex, year of ART initiation, occupation type, marital status, education status, heterosexual sex as a risk factor, WHO stage, tuberculosis at entry, and viral load using a random effects Cox proportional hazards model that controlled for site variability, we found that the associations between CD4+ cell count and LTFU as well as VL and LTFU, in a
4

time-dependent manner, remained statistically significant (Table 2; P < 0.001). Interestingly, the associations between sex and LTFU as well as regimen and LTFU did not remain following addition of CD4+ count and VL as time-dependent variables. Further, although the statistical significance of the associations between ART initiation year and WHO stage, respectively, with LTFU changed upon imputation of missing data, the associations between CD4+ count with LTFU and VL with LTFU, respectively, were not affected.
Early Adherence Patterns Predict LTFU by Month 12 Post-Initiation of ART

Of the patients with at least 3 months on treatment, over half were 100% adherent during the first 3 months. As the largest percentage of loss typically occurred during the first 12 months of treatment, we conducted a subset analysis on the 47 656 patients with at least 3 months on ART to examine the association between adherence patterns during the first 3 months of ART and LTFU by month 12. In adjusted multivariate logistic regression analyses, controlling for site variability, age, sex, year of ART initiation, occupation, marital status, education, baseline WHO stage, baseline CD4+ cell count, baseline VL, and initial ART regimen, the association between adherence and LTFU remained, with a trend of reduced risk of LTFU with better adherence during the first 3 months of treatment. Specifically,

•

OFID

•

Meloni et al

Table 1.

Demographics and Baseline Clinical Characteristics of ARV-Naive Treatment Cohort
(b) Patients discontinued after only 1 pick-up 5423 0 35 (29–41) 3130 (57.7) 2293 (42.3) 549 (10.1) 4874 (89.9) 386 (7.1) 431 (7.9) 1027 (18.9) 1297 (23.9) 1124 (20.7) 1158 (21.4) 1155 (22.5) 1179 (23.0) 1830 (35.6) 970 (18.9) 1127 (21.7) 2872 (55.4) 558 (10.8) 630 (12.2) 1.279 (24.7) 3628 (70.1) 268 (5.2) 4596 (94.9) 247 (5.1) (c) Patients retained after 1st pick-up 51 953 28.6 (14.8–44.9) 35 (29–41) 33 676 (64.8) 18 277 (35.2) 5183 (10.0) 46 770 (90.0) 4428 (8.5) 4714 (9.1) 6945 (13.4) 11 754 (22.6) 11 714 (22.5) 12 398 (23.9) 9421 (18.5) 11 194 (21.9) 18 025 (32.3) 12 436 (24.3) 10 084 (19.6) 29 697 (57.7) 4559 (8.9) 7142 (13.9) 12 612 (24.6) 35 458 (69.0) 3292 (6.4) 46 273 (95.5) 2188 (4.5) 0.063 0.002 <0.001 <0.001 (d) P-value for b vs c – <0.0001 0.49 <0.001 (e) Patients LTFU after 2nd pick-up 14 626 9.3 (2.9–21.0) 34 (28–41) 8861 (60.6) 5765 (39.4) 1108 (7.6) 13 518 (92.4) 1333 (9.1) 1756 (12.0) 2424 (16.6) 3545 (24.2) 3034 (20.7) 2534 (17.3) 3160 (22.1) 3292 (23.0) 4910 (34.3) 2966 (20.7) 3219 (22.3) 7814 (54.2) 1445 (10.0) 1945 (13.5) 3890 (27.0) 9747 (67.7) 757 (5.3) 13 031 (95.8) 574 (4.2) ( f ) Patients retained after 2nd pick-up 34 996 36.6 (23.9–51.2) 35 (29–41) 23 307 (66.6) 11 689 (33.4) 3705 (10.6) 31 291 (89.4) 2837 (8.1) 2630 (7.5) 4130 (11.8) 7706 (22.0) 8183 (23.4) 9510 (27.2) 5811 (16.9) 7422 (21.5) 12 344 (35.8) 8896 (25.8) 6398 (18.4) 20 584 (59.2) 2902 (8.4) 4868 (14.0) 8108 (23.4) 24 178 (69.8) 2377 (6.9) 31 152 (95.3) 1532 (4.7) 0.028 <0.001 <0.001 <0.001 (g) P-value for e vs f – 0.0001 <0.0001 <0.001

Characteristic N Median time on ART, months (IQR) Median age, in years (IQR) Sex, n (%) Female Male Site Type Secondary Tertiary ART Initiation Year Jun 2004–Dec 2005 2006 2007 2008 2009 2010–Feb 2011 Education, n (%) None Primary Secondary Tertiary Marital Status, n (%) Single Married Divorced/Separated Widowed Occupation Non-income-generating Laborer/Service Worker Manager/Professional HIV Risk Factor Heterosexual Sex Other/multiple Time-Dependent Predictors of LTFU

(a) All patients 57 376 25.7 (10.8–43.1) 35 (29–41) 36 806 (64.1) 20 570 (35.9) 5732 (10.0) 51 644 (90.0) 4814 (8.4) 5145 (9.0) 7972 (13.9) 13 051 (22.7) 12 838 (22.4) 13 556 (23.6) 10 576 (18.8) 12 373 (22.0) 19 855 (35.3) 13 406 (23.9) 11 211 (19.8) 32 569 (57.5) 1446 (10.0) 1946 (13.5) 13 891 (24.6) 39 086 (69.1) 3560 (6.3) 50 869 (95.4) 2435 (4.6)

0.73

>0.001

<0.001

<0.001

•
OFID

•
5

6

•
OFID

Table 1 continued.
(b) Patients discontinued after only 1 pick-up 1223 (26.5) 977 (21.2) 1386 (30.1) 1026 (22.2) 340 (8.2) 489 (11.8) 1267 (30.7) 2033 (49.2) 614 (16.5) 882 (23.7) 1362 (36.7) 856 (23.1) 1037 (19.1) 4386 (80.9) 608 (17.3) 2912 (982.7) 163 (4.7) 3317 (95.3) 1245 (23.0) 1124 (20.7) 536 (9.9) 2035 (37.5) 54 (1.0) 429 (7.9) (c) Patients retained after 1st pick-up 7715 (15.6) 8766 (17.7) 19 054 (38.5) 14 012 (28.3) 3879 (8.6) 6261 (13.9) 15 675 (34.8) 19 221 (42.7) 10 919 (24.3) 12 858 (26.7) 15 028 (33.5) 6031 (13.5) 11 492 (22.1) 40 461 (77.9) 5575 (15.8) 29 624 (84.2) 1980 (5.7) 32 921 (94.3) 9097 (17.5) 10 265 (19.8) 4783 (9.2) 24 147 (46.5) 248 (0.5) 3413 (6.6) <0.001 0.015 0.025 <0.001 <0.001 <0.001 (d) P-value for b vs c <0.001 (e) Patients LTFU after 2nd pick-up 2718 (19.6) 2685 (19.4) 5084 (36.7) 3359 (24.3) 1083 (8.6) 1653 (13.1) 4229 (33.5) 5671 (44.9) 2412 (19.8) 3150 (25.9) 4321 (35.5) 2300 (18.9) 3484 (23.8) 11 142 (76.2) 1828 (17.5) 8644 (82.5) 577 (5.6) 9805 (94.4) 2683 (18.3) 2977 (20.4) 1570 (10.7) 6230 (42.6) 93 (0.6) 1073 (7.3) ( f ) Patients retained after 2nd pick-up 4544 (13.6) 5586 (16.7) 13 193 (39.4) 10 143 (30.3) 2633 (8.7) 4343 (14.3) 10 761 (35.4) 12 645 (41.6) 8046 (26.3) 9108 (29.7) 10 018 (32.7) 3452 (11.3) 27 595 (78.9) 7401 (21.1) 3497 (15.2) 19 592 (84.8) 1247 (5.5) 21 647 (94.5) 6045 (17.3) 6758 (19.3) 2989 (8.5) 16 894 (48.3) 141 (0.4) 2169 (6.2) <0.001 0.54 <0.001 <0.001 <0.001 <0.001 (g) P-value for e vs f <0.001

•
Meloni et al

Characteristic Baseline CD4 count, cells/mL ≤50 51–100 101–200 >200 Baseline viral load, copies/mL 0–999 1000–9999 10 000–99 999 ≥100 000 WHO Stage, n (%) 1 2 3 4 TB at Entry Yes No HBV Status at Baseline Positive Negative HCV Status at Baseline Positive Negative First-line drug regimen TDF + XTC + EFV TDF + XTC + NVP AZT + 3TC + EFV AZT + 3TC + NVP d4T + 3TC + EFV d4T + 3TC + NPV

(a) All patients 8938 (16.5) 9743 (18.0) 20 440 (37.7) 15 038 (27.8) 4219 (8.6) 6750 (13.7) 16 942 (34.5) 21 254 (43.2) 11 533 (23.7) 13 740 (28.3) 16 390 (33.8) 6887 (14.2) 12 529 (21.8) 44 847 (78.2) 6183 (16.0) 32 536 (84.0) 2143 (5.6) 36 238 (94.4) 10 342 (18.0) 11 389 (19.9) 5319 (9.3) 26 182 (45.6) 302 (0.5) 3842 (6.7)

Abbreviations: ART, antiretroviral therapy; ARV, antiretroviral; HBV, hepatitis B virus; HCV, hepatitis C virus; HIV, human immunodeficiency virus; IQR, interquartile range; TB, tuberculosis; WHO, World Health Organization.

Fig. 2. CD4+ cell counts and viral loads predict LTFU in time-dependent manner. Abbreviations: ART, antiretroviral therapy; LTFU, loss to follow-up; VL, viral load.

patients with 50%–94% adherence were at 32% lower risk (95% confidence interval [CI]: 0.61–0.76), those with 94%–99% adherence were at 46% lower risk (95% CI: 0.48–0.61), and those with 100% adherence were at 64% lower risk (95% CI: 0.32–0.40) of being LTFU by month 12, as compared to those patients with <50% adherence during the first 3 months of treatment (Fig. 3). DISCUSSION To our knowledge, this evaluation is the first to examine multiple time-dependent predictors of LTFU using nearly 7.5 years of electronically maintained patient-level data on nearly 52 000 patients. The data were captured starting at the initiation of a rapidly scaled-up HIV treatment program. Over the course of 7 years, the activities were decentralized, moving outward from tertiary to secondary level centers. Because data were collected from program initiation, we were able to compare LTFU rates from patients enrolled as early as 2004 to those entering in 2011. Overall retention rates in the Harvard/APIN PEPFAR program were comparable to those reported in other studies [4,27,28]. Similar to other studies, we also found that the majority of loss occurs within the first 12–18 months of treatment [27–31]. Interestingly, in complete case analyses, we found that LTFU rates were generally lower for those enrolled prior to 2006 than after 2006. We hypothesize that LTFU might have increased with calendar time due to the expanding nature of the scale-up program and decentralization of care, with the provision of services being shifted from tertiary-level sites down to secondary and primary sites (ie, scale-up effect). In addition, with other programs also offering care and opening additional sites, it is possible that patients moved to sites closer to

their homes. Because there was no existing mechanism to independently track movements of patients between sites in Nigeria, we were unable to track transfers to sites outside the Harvard/ APIN PEPFAR network. Thus, part of the decreased risk of loss associated with calendar time might be due to undocumented transfers. It is noteworthy, however, that the association between calendar time and LTFU did not remain following imputation to time-dependent CD4+ count and VL data. This study identified some important predictors of LTFU, particularly those that remain continuous predictors of outcomes over time through 4 years of observation. Because the program included such a large study population, we were overpowered to find statistical significance in the smallest difference. However, we found that some of the differences were of notable magnitude. Other studies have also identified lower age, marital status, and lower baseline CD4+ cell count as predictors of LTFU [7,8,27,30,32–37]. But to our knowledge, this is the first study to show that CD4+ cell count and VL suppression rates remain predictors of LTFU in a time-dependent fashion. A few prior studies have found a correlation with adherence patterns and overall survival [11,22]. Our analysis was unique in that we focused on early adherence pattern as a predictor of future outcomes to show time-dependent effects. This study has several strengths based on its evaluation of a large HIV treatment program. First, the study had a large sample size with nearly 52 000 patients. Second, the evaluation used electronically stored patient-level data collected at 32 hospitals and clinics across Nigeria, thus making the results of the evaluation more generalizable. The data were collected starting in 2004 through February 2012, allowing for monitoring of temporal trends in LTFU and predictors over a significant period. Additionally, with over 7.5 years of data, we were able to examine rates and predictors of long-term retention. Furthermore,
Time-Dependent Predictors of LTFU

•

OFID

•

7

Table 2. Cox Proportional Hazards Model of Time-Independent and Time-Dependent Predictors of LTFU among ART-naive Adult Patients That Made at Least 2 ART Pick-ups (n = 51 953)
Adjusted cox values complete casesa aHR Ref 0.81 0.75 0.66 Ref 0.98 Ref 1.24 1.74 1.52 1.89 2.12 95% CI – 0.66–0.98 0.59–0.95 0.50–0.86 – 0.87–1.10 – 1.11–1.37 1.34–2.27 1.20–1.93 1.60–2.23 1.58–2.86 P-value – 0.036 0.015 0.003 – 0.69 – <0.001 <0.001 0.001 <0.001 <0.001 Adjusted cox values multiple imputationsa aHR Ref 0.8 0.76 0.67 Ref 0.91 Ref 1.12 1.24 1.03 1 0.84 95% CI – 0.70–0.91 0.66–0.88 0.55–0.82 – 0.80–1.03 – 1.01–1.25 0.96–1.60 0.73–1.45 0.74–1.34 0.49–1.43 P-value – 0.001 <0.001 <0.001 – 0.14 – 0.04 0.11 0.87 0.98 0.52

Unadjusted cox values Variable Age, years <30 30–34 35–40 >40 Sex Male Female ART Initiation Year Jun 2004–Dec 2005 2006 2007 2008 2009 2010–Feb 2011 Site Type Secondary Tertiary Occupation Non-income-generating Laborer/Service Worker Manager/Professional Marital Status, n (%) Single/Divorced/Separated Married/Widowed Education, n (%) None Primary Secondary Tertiary HIV Risk Factor Heterosexual Sex Other/multiple WHO Stage 1 2 3 4 TB at Entry HBV at Entry HCV at Entry CD4+ Cell Count, cells/mL ≤50 51–100 101–200 >200 Undetectable Viral Load Ref 1.11 1.34 1.82 1.1 1.11 0.94 Ref 0.46 0.25 0.13 0.31 – 1.05–1.17 1.28–1.41 1.72–1.92 1.06–1.14 1.05–1.16 0.86–1.02 – 0.38–0.57 0.21–0.30 0.11–0.15 0.28–0.34 – <0.001 <0.001 <0.001 <0.001 <0.001 0.15 – <0.001 <0.001 <0.001 <0.001 Ref 1.18 – 1.02–1.37 – 0.029 Ref 0.9 0.7 Ref 0.8 Ref 0.86 0.78 0.65 – 0.87–0.94 0.65–0.76 – 0.77–0.83 – 0.82–0.90 0.75–0.82 0.62–0.68 – <0.001 <0.001 – <0.001 – <0.001 <0.001 <0.001 Ref 1.14 – 1.07–1.21 – <0.001 Ref 0.86 0.83 0.84 Ref 0.82 Ref 1.38 1.41 1.31 1.26 1.23 – 0.82–0.90 0.79–0.87 0.81–0.88 – 0.79–0.85 – 1.34–1.42 1.37–1.45 1.28–1.34 1.23–1.30 1.20–1.27 – <0.001 <0.001 <0.001 – <0.001 – <0.001 <0.001 <0.001 <0.001 <0.001 HR 95% CI P-value

Ref 1.1 1.23 Ref 0.82 Ref 0.77 0.76 0.65 Ref 1.58 Ref 1.24 1.24 1.29 0.97

– 1.01–1.20 0.97–1.57 – 0.72–0.94 – 0.67–0.88 0.67–0.86 0.56–0.76 – 0.48–5.21 – 0.99–1.55 1.01–1.51 0.85–1.95 0.81–1.16

– 0.021 0.093 – 0.004 – <0.001 <0.001 <0.001 – 0.45 – 0.06 0.04 0.23 0.77

Ref 1.02 1.22 Ref 0.76 Ref 0.82 0.81 0.71 Ref 1.17 Ref 1.14 1.21 1.38 0.95

– 0.94–1.11 0.99–1.51 – 0.70–0.82 – 0.72–0.93 0.75–0.88 0.57–0.89 – 0.78–1.75 – 0.94–1.37 1.04–1.40 1.07–1.79 0.83–1.08

– 0.58 0.07 – <0.001 – 0.002 <0.001 0.003 – 0.44 – 0.2 0.013 0.013 0.39

Ref 0.56 0.34 0.19 0.4

– 0.47–0.68 0.25–0.47 0.16–0.24 0.34–0.49

– <0.001 <0.001 <0.001 <0.001

Ref 0.64 0.46 0.25 0.4

– 0.50–0.82 0.32–0.66 0.19–0.33 0.35–0.46

– <0.001 <0.001 <0.001 <0.001

8

•

OFID

•

Meloni et al

Table 2 continued.
Adjusted cox values complete casesa aHR Ref 0.86 0.98 0.89 1.01 0.73 95% CI – 0.63–1.17 0.77–1.25 0.66–1.22 0.74–1.38 0.37–1.42 P-value – 0.34 0.86 0.48 0.97 0.35 Adjusted cox values multiple imputationsa aHR Ref 0.88 0.97 0.9 0.98 0.83 95% CI – 0.69–1.13 0.79–1.20 0.75–1.08 0.79–1.21 0.60–1.15 P-value – 0.31 0.81 0.26 0.85 0.26

Unadjusted cox values Variable Starting ART Regimen TDF+FTC/3TC + EFV TDF + FTC/3TC + NVP AZT + 3TC + EFV AZT + 3TC + NVP d4T + 3TC + EFV d4T + 3TC + NVP Ref 0.92 1 0.79 1.07 0.76 – 0.87–0.97 0.94–1.06 0.76–0.83 0.87–1.31 0.71–0.81 – 0.002 0.95 <0.001 0.54 <0.001 HR 95% CI P-value

Abbreviations: ART, antiretroviral therapy; CI, confidence interval; EFV, efavirenz; FTC,emtricitabine; HBV, hepatitis B virus; HCV, hepatitis C virus; HIV, human immunodeficiency virus; TB, tuberculosis; TDF,tenofovir; 3TC, lamivudine; WHO, World Health Organization.
a

Random effects Cox proportional hazards model generated to control for site variability.

because laboratory data were electronically collected on a patient level at 6-month intervals, we were able to look at timedependent variables as predictors of LTFU on an individual patient level.

The study was limited because the program did not actively trace all patients that were lost, which is not atypical from other ART programs. Our study is also limited in that we did not conduct retrospective analyses on data regarding reasons for

Fig. 3. Results from random effects logistic regression model with multiple imputations examining association between early adherence patterns and LTFU by M12 post-initiation of ART. Abbreviations: ART, antiretroviral therapy; LTFU, loss to follow-up; M12, month 12.
Time-Dependent Predictors of LTFU

•

OFID

•

9

discontinuation. If we were able to trace and administer surveys that solicited additional data from LTFU patients, we would have more robust information for our clinics to improve their patient retention rates; we anticipate that having better information on reasons for loss would subsequently lower our LTFU rates and potentially affect the magnitudes of associations regarding predictors of loss. Finally, because this analysis focused only on ART patients and we were looking only at information from ART initiation through time on ART, we were not looking at additional predictors from the pre-ART phase that also might have explained retention patterns. Studies that traced lost patients found that up to 50% of those cumulatively lost had actually died [5,13,38,39] and that most deaths occurred within 30 days of the last clinical encounter with the patient [5]. Other studies that tracked lost patients found that some simply moved to other health facilities or chose to take a break from treatment due to insufficient funds to attend clinic, food insecurity, difficulty procuring childcare, fear of stigma, or issues with adverse effects [4,39,40]. As such, by using the LTFU composite outcome, and due to the fact that the program only passively collects death and transfer information, we are underestimating those that have died or left the program. Various researchers have shown that developing interventions to address specific barriers can readily address the problems and encourage some patients to return [4]. It is our belief that using factors predictive of loss for targeted interventions before a patient is LTFU will be particularly helpful, specifically for those patients that are lost due to reasons other than death. For example, understanding that early adherence patterns strongly correlate with future LTFU could serve as an easy trigger for targeted adherence counseling. Furthermore, knowing that lower relative CD4+ counts or unsuppressed VL, at any time point in treatment, is predictive of loss can also serve as a powerful and simple tool for targeted interventions. In summary, between 2004 and 2012, we found that a significant proportion of patients enrolled in the Harvard/APIN PEPFAR treatment program were eventually LTFU. Understanding that CD4+ cell counts, VLs and early adherence patterns are strong predictors of future loss will aid ART programs in identifying patients for targeted interventions to improve retention rates.
Acknowledgements
The authors gratefully acknowledge the patients and the incredible work and dedication of the clinical, data, and laboratory staff at all of the Harvard/ APIN PEPFAR ART sites. We also wish to thank Dr Eric Tchetgen Tchetgen of the Harvard School of Public Health, Department of Biostatistics, and Dr Geoffrey Eisen, of the Northwestern School of Medicine, for their feedback on the statistical methods used in these evaluations.

(U51HA02522) and the Centers for Disease Control and Prevention (CDC) through a cooperative agreement with APIN (PS 001058). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. Conflict of Interest Statement: The authors report no conflicts of interest.

References
1. Federal Ministry of Health Nigeria. 2010 National HIV sero-prevalence sentinel survey. Abuja, Nigeria, 2010. 2. Federal Republic of Nigeria National Agency for the Control of AIDS. Global AIDS response: Country progress report. Nigeria, 2012. 3. World Bank. Nigeria: Country at a glance. Washington, DC The World Bank, 2012. 4. Rosen S, Fox MP, Gill CJ. Patient retention in antiretroviral therapy programs in sub-Saharan Africa: a systematic review. PLoS Med. 2007;4 (10):e298. 5. Dalal RP, Macphail C, Mqhayi M, et al. Characteristics and outcomes of adult patients lost to follow-up at an antiretroviral treatment clinic in Johannesburg, South Africa. J Acquir Immune Defic Syndr. 2008;47 (1):101–107. 6. Ahonkhai AA, Noubary F, Munro A, et al. Not all are lost: Interrupted laboratory monitoring, early death, and loss to follow-up (LTFU) in a large South African treatment program. PLoS One. 2012;7(3):e32993. 7. Somi G, Keogh SC, Todd J, et al. Low mortality risk but high loss to follow-up among patients in the Tanzanian national HIV care and treatment programme. Trop Med Int Health. 2012;17(4): 497–506. 8. Zhou J, Tanuma J, Chaiwarith R, et al. Loss to follow-up in HIV-infected patients from Asia-Pacific Region: Results from TAHOD. AIDS Res Treat. 2012;2012:375217. 9. Alvarez-Uria G, Naik PK, Pakam R, et al. Factors associated with attrition, mortality, and loss to follow up after antiretroviral therapy initiation: Data from an HIV cohort study in India. Glob Health Action. 2013;6:21682. 10. Clouse K, Pettifor AE, Maskew M, et al. Patient retention from HIV diagnosis through one year on antiretroviral therapy at a primary health care clinic in Johannesburg, South Africa. J Acquir Immune Defic Syndr. 2013;62(2):e39–e46. 11. Karcher H, Omondi A, Odera J, et al. Risk factors for treatment denial and loss to follow-up in an antiretroviral treatment cohort in Kenya. Trop Med Int Health. 2007;12(5):687–694. 12. Brinkhof MW, Dabis F, Myer L, et al. Early loss of HIV-infected patients on potent antiretroviral therapy programmes in lower-income countries. Bull World Health Organ. 2008;86(7):559–567. 13. Geng EH, Glidden DV, Emenyonu N, et al. Tracking a sample of patients lost to follow-up has a major impact on understanding determinants of survival in HIV-infected patients on antiretroviral therapy in Africa. Trop Med Int Health. 2010;15(Suppl 1):63–69. 14. Krishnan S, Wu K, Smurzynski M, et al. Incidence rate of and factors associated with loss to follow-up in a longitudinal cohort of antiretroviral-treated HIV-infected persons: An AIDS Clinical Trials Group (ACTG) Longitudinal Linked Randomized Trials (ALLRT) analysis. HIV Clin Trials. 2011;12(4):190–200. 15. Mosoko JJ, Akam W, Weidle PJ, et al. Retention in an antiretroviral therapy programme during an era of decreasing drug cost in Limbe, Cameroon. J Int AIDS Soc. 2011;14:32. 16. Wang B, Losina E, Stark R, et al. Loss to follow-up in a community clinic in South Africa–roles of gender, pregnancy and CD4 count. S Afr Med J. 2011;101(4):253–257. 17. Charurat M, Oyegunle M, Benjamin R, et al. Patient retention and adherence to antiretrovirals in a large antiretroviral therapy program in Nigeria: a longitudinal analysis for risk factors. PLoS One. 2010;5(5): e10584. 18. Odafe S, Torpey K, Khamofu H, et al. The pattern of attrition from an antiretroviral treatment program in Nigeria. PLoS One. 2012;7(12): e51254.

Funding
This work was funded, in part, by the US Department of Health and Human Services, Health Resources and Services Administration

10

•

OFID

•

Meloni et al

19. World Health Organization. Antiretroviral therapy for HIV infection in adults and adolescents: recommendations for a public health approach—2006 rev. Geneva: World Health Organization, 2006. 20. World Health Organization. Antiretroviral therapy for HIV infection in adults and adolescents: recommendations for a public health approach—2010 rev. Geneva: WHO, 2010. 21. Grossberg R, Zhang Y, Gross R. A time-to-prescription-refill measure of antiretroviral adherence predicted changes in viral load in HIV. J Clin Epidemiol. 2004;57(10):1107–1110. 22. Nachega JB, Hislop M, Dowdy DW, et al. Adherence to highly active antiretroviral therapy assessed by pharmacy claims predicts survival in HIV-infected South African adults. J Acquir Immune Defic Syndr. 2006;43(1):78–84. 23. Grossberg R, Gross R. Use of pharmacy refill data as a measure of antiretroviral adherence. Curr HIV/AIDS Rep. 2007;4(4):187–191. 24. Seguy N, Diaz T, Campos DP, et al. Evaluation of the consistency of refills for antiretroviral medications in two hospitals in the state of Rio de Janeiro, Brazil. AIDS Care. 2007;19(5):617–625. 25. Bisson GP, Gross R, Bellamy S, et al. Pharmacy refill adherence compared with CD4 count changes for monitoring HIV-infected adults on antiretroviral therapy. PLoS Med. 2008;5(5):e109. 26. El-Khatib Z, Katzenstein D, Marrone G, et al. Adherence to drug-refill is a useful early warning indicator of virologic and immunologic failure among HIV patients on first-line ART in South Africa. PLoS One. 2011;6(3):e17518. 27. Ochieng-Ooko V, Ochieng D, Sidle JE, et al. Influence of gender on loss to follow-up in a large HIV treatment programme in western Kenya. Bull World Health Organ. 2010;88(9):681–688. 28. Ekouevi DK, Balestre E, Ba-Gomis FO, et al. Low retention of HIVinfected patients on antiretroviral therapy in 11 clinical centres in West Africa. Trop Med Int Health. 2010;15(Suppl 1):34–42. 29. Rasschaert F, Koole O, Zachariah R, et al. Short and long term retention in antiretroviral care in health facilities in rural Malawi and Zimbabwe. BMC Health Serv Res. 2012;12:444. 30. Maru DS, Khakha DC, Tahir M, et al. Poor follow-up rates at a self-pay northern Indian tertiary AIDS clinic. Int J Equity Health. 2007;6:14.

31. Cicconi P, Cozzi-Lepri A, Castagna A, et al. Insights into reasons for discontinuation according to year of starting first regimen of highly active antiretroviral therapy in a cohort of antiretroviral-naive patients. HIV Med. 2010;11(2):104–113. 32. Mocroft A, Kirk O, Aldins P, et al. Loss to follow-up in an international, multicentre observational study. HIV Med. 2008;9(5):261–269. 33. Bekolo CE, Webster J, Batenganya M, et al. Trends in mortality and loss to follow-up in HIV care at the Nkongsamba Regional hospital, Cameroon. BMC Res Notes. 2013;6(1):512. 34. Guy R, Wand H, McManus H, et al. Antiretroviral treatment interruption and loss to follow-up in two HIV cohorts in Australia and Asia: implications for ‘test and treat’ prevention strategy. AIDS Patient Care STDS. 2013;27(12):681–691. 35. Honge BL, Jespersen S, Nordentoft PB, et al. Loss to follow-up occurs at all stages in the diagnostic and follow-up period among HIV-infected patients in Guinea-Bissau: A 7-year retrospective cohort study. BMJ Open. 2013;3(10):e003499. 36. Rich ML, Miller AC, Niyigena P, et al. Excellent clinical outcomes and high retention in care among adults in a community-based HIV treatment program in rural Rwanda. J Acquir Immune Defic Syndr. 2012;59 (3):e35–e42. 37. Mutasa-Apollo T, Shiraishi RW, Takarinda KC, et al. Patient retention, clinical outcomes and attrition-associated factors of HIV-infected patients enrolled in Zimbabwe’s national antiretroviral therapy programme, 2007–2010. PLoS One. 2014;9(1):e86305. 38. Geng EH, Emenyonu N, Bwana MB, et al. Sampling-based approach to determining outcomes of patients lost to follow-up in antiretroviral therapy scale-up programs in Africa. JAMA. 2008;300(5):506–507. 39. Geng EH, Bangsberg DR, Musinguzi N, et al. Understanding reasons for and outcomes of patients lost to follow-up in antiretroviral therapy programs in Africa through a sampling-based approach. J Acquir Immune Defic Syndr. 2010;53(3):405–411. 40. Tuller DM, Bangsberg DR, Senkungu J, et al. Transportation costs impede sustained adherence and access to HAART in a clinic population in southwestern Uganda: A qualitative study. AIDS Behav. 2010;14 (4):778–784.

Time-Dependent Predictors of LTFU

•

OFID

•

11