In Mozambique, a country with a high burden of HIV, more than 170,000 people were receiving antiretroviral therapy (ART) at the end of 2009, a dramatic increase from the 7000 receiving ART in 2004.1,2 As the number of people receiving ART has rapidly increased, more strain has been placed on the health system and its workforce. Larger patient populations can bottleneck the existing infrastructure and overwhelm the clinicians and pharmacy staff providing services. An overstrained health system could result in longer wait times and shorter lower quality patient-provider interactions, leading to patient dissatisfaction3,4 and potentially higher attrition from ART programs.
Many view inadequate human resource capacity as one of the biggest obstacles with regard to effective ART delivery in resource-limited settings.5-15 Mozambique faces one of the lowest provider-to-population ratios in the world with 3 doctors, 21 nurses, and 3 pharmacy staff per 100,000 people.12 Understanding the influence of patient volume and health workforce levels on retention in ART programs in resource-poor settings is crucial for the effective delivery of HIV treatment services.
We conducted a retrospective cohort study to examine the association of patient volume, clinical staff burden, pharmacy staff burden, and patient characteristics with attrition from ART programs in Manica and Sofala provinces of Mozambique.
At the end of 2004, central Mozambique had one of the heaviest HIV burdens in the country with an estimated 20.4% of the adult population infected with HIV.16 The first ART delivery site in the 2 provinces opened in Sofala's Beira Central Hospital in 2003 although a second one opened in Manica's Chimoio Provincial Hospital in 2004. In 2006, the Ministry of Health integrated ART services into primary health care clinics distributed throughout 23 districts in Manica and Sofala provinces to improve access to HIV care and treatment, with primary funding through the President's Emergency Plan for AIDS Relief.
Study subjects were adult (≥15 years of age), ART-naive patients who initiated ART between January 1, 2006, and June 1, 2008, in 18 public sector clinics, which had electronic clinic databases in Sofala or Manica province. For 2 study clinics, we excluded patients enrolling into treatment after December 31, 2007, and March 1, 2008, due to 2 distinct historical events-flooding and temporary closure of the clinic, respectively. Patients who were pregnant at enrollment were excluded from the study.
This study utilized electronic clinic databases, which contained sociodemographic characteristics, pharmacy and clinical data, and human resources information. These databases were primarily used by clinic management as tools for routine clinic monitoring and evaluation. Numerous range checks and cross checks were regularly applied to the databases to minimize data errors. Database managers performed queries to monitor data quality and provided on-the-job training to technicians if discrepancies arose. Additional data about personnel working at the ART sites was collected from routine health systems data supplemented with recording systems from the nongovernmental organization, Health Alliance International, which provided technical support to the sites. The databases were collected in May of 2009, which provided approximately 11 months for peer counselors to track patients who may have missed appointments toward the end of the study. This approach made outcome ascertainment and data transcription more comparable throughout the study period.
To evaluate the reliability of data used in our study, we compared data from the clinic databases to a randomly selected sample of 520 paper-based patient medical records from the clinics that were included in the study. Key variables were found to agree between 92.6% and 99.2% of the time with kappa scores ranging from 0.91 to 0.97, indicating “almost perfect” agreement.17
Patient Volume and Human Resource Levels
The primary clinic-level exposures of interest were the monthly number of clinical and pharmacy visits (patient volume), the monthly number of clinical visits per clinician (Clinical Staff Burden) and the monthly number of pharmacy visits per pharmacy staff (Pharmacy Staff Burden). The number of clinicians, including physicians, nurses, and clinical officers, was determined from the clinic databases. Clinicians who saw at least 1 patient on a particular day were classified as providing 1 working day. Regardless of whether a clinician saw 1 patient or 10 patients on a given day, he or she would be classified as contributing 1 working day. For a particular month, the number of working days provided by clinicians were aggregated and divided by the number of days the clinic was open to obtain the monthly average number of clinicians treating patients. We combined physicians, nurses, and clinical officers into 1 clinician category due, in part, to previous research showing similar levels of quality of care delivered by physicians and nonphysician clinicians.18 The number of pharmacy staff working each month and whether they worked on a part-time or full-time basis was abstracted from the routine, nongovernmental organization health systems data. The number of clinic visits and the number of pharmacy visits were then divided by the estimated number of clinical staff and pharmacy staff working for a given month, respectively.
Patient volume, clinical staff burden, and pharmacy staff burden were treated as time-varying exposures over 6-month intervals beginning with when a site initiated ART delivery. For each interval, we took the average of the monthly patient volume, number of clinical visits per clinician, and number of pharmacy visits per pharmacy staff as the estimates. The 6-month time interval was chosen to allow enough follow-up time for outcomes to occur for a given exposure level. Pharmacy staffing data were unavailable in the routine health systems data for the study period for 1 clinic but was available for the remaining 17 clinics. Other clinic-level covariates included the clinic model of care—whether the clinic operated as a stand-alone HIV care and treatment clinic (vertical) or was integrated within primary health centers (integrated); the clinic location—whether the clinic was in a rural or urban setting; and clinic experience—whether the clinic was in its first 6 months of offering HIV treatment or after its first 6 months of offering treatment. Patient-level data were also provided by the clinic databases.
If a patient failed to return for ART medication refill, “activistas” (clinic-based peer counselors) were notified and begin actively tracing patients. The peer counselors visited the patient's residence and encouraged them to return for treatment at the clinic if the patient was still alive. If the patient had died, the date of death was documented. Outcomes for patients who initiated HIV treatment were defined as (1) transferred—patients who transferred to receive treatment at another facility; (2) suspended—a patient who suspended HIV treatment per clinician's recommendation but remained in care; (3) dead—a patient who died due to any cause while receiving ART which was ascertained by active tracing of patients who missed a pharmacy visit and reports from family members who informed the clinic of the patient's death; and (4) lost to follow-up—a patient who had failed to return for treatment for 2 months beyond their missed ART refill visit for reasons unrelated to patient death. Patients who continued to receive ART were considered to be maintained on ART until they either experienced one of the terminating events or reached the end of study. For patients who met the definition of lost to follow-up, but then decided to reinitiate treatment at a later point in time, they were still considered lost to follow-up and their person-time after reinitiating treatment would be excluded from the study because one of the inclusion criteria specifies that patients are ART naive.
We defined our outcome measure of attrition from care as patients classified as either dead or lost to follow-up. One of the principal reasons for combining the 2 outcomes was that differential ascertainment between those patients who were lost to follow-up and those who died could occur at clinics with larger patient volumes. Essentially, the workload for peer counselors could have compromised their ability to determine whether or not patients who had failed to return for a pharmacy refill actually died, and therefore, they could be incorrectly classified as lost to follow-up. The date of attrition was defined as the last pharmacy refill date for those who were lost to follow-up and the date of death for those who died.
Cox proportional hazards models were used to analyze time until attrition, defined as death or lost to follow-up. Individuals who were categorized as transferred or suspended were treated as censored observations at the time of transfer or suspension. Follow-up time began at initiation of HIV treatment and ended with the date of death, transfer, treatment suspension, loss to follow-up (date of last clinic or pharmacy visit), or June 30, 2008, whichever came first. Patients attending the 2 clinics with historical events were censored on dates corresponding to the event—December 31, 2007, and March 1, 2008. Given that survival times were clustered at the clinic-level, robust variances were used.19
The primary analysis tested the association of patient volume, clinical staff burden, and pharmacy staff burden with time until attrition. We modeled these exposures as 3-level categorical variables, split at the tertile of the distribution (patient volume—low: <393, medium: 393 to <982, high: ≥982; clinical staff burden—low: <77, medium: 77 to <157, high: ≥157; pharmacy staff burden—low: <164, medium: 164 to <539, high: ≥539). Each of these analyses adjusted for the following a priori specified potential confounders: pre-ART CD4 count, pre-ART WHO stage, age at enrollment at the ART clinic, education, gender, and clinic model of care. We evaluated whether the associations of patient-level characteristics with patient attrition varied between early follow-up (≤4 months) and later follow-up (>4 months).20 Patient characteristics with an interaction P value less than 0.2 between early and later follow-up were retained in the models. We used residual plots to guide the choice of the best-fitting transformation for continuous variables. In exploratory analysis, we evaluated the impact of adjusting for clinic experience, location of the clinic and calendar time in our models. Variables that changed the coefficient estimates for the primary exposures by ≥10% were considered important confounders and retained in the model.
Data were missing for some patients on CD4 count (5%), education (7%), and WHO stage (9%). To address this, we utilized multiple imputation procedures to fill in missing values in the dataset. Ten imputations were done, using fully conditional specifications including pre-ART CD4 count, pre-ART WHO stage, age at enrollment, gender, education, attrition status, and survival time (log form).21 Results from the imputations were combined into a single set of parameter estimates for the “final” proportional hazards regression model that incorporated the uncertainty from the imputations.21,22
The study was approved by the institutional review boards of the Mozambique Ministry of Health and the University of Washington. All analyses were conducted in Stata version 11.1 (College Station, TX).
A total of 18 of the 36 HIV care and treatment clinics recognized by the Ministry of Health from Manica and Sofala provinces were included in the analysis. The remaining 18 were not included because they did not have electronic patient tracking systems. Table 1 describes characteristics of these clinics. During the study period, 1 clinic shifted from a vertical model of HIV treatment delivery to a model integrated with primary health care, resulting in 3 clinics providing patient follow-up time under a vertical model and 16 clinics providing patient follow-up time under an integrated model.
Of the 15,232 patients receiving ART at the 18 study clinics between January 1, 2006, and June 1, 2008, we excluded 810 who were transferred in from another facility, 1138 who were under 15 years of age and 1491 who were pregnant at the time of enrollment. Thus, 11,793 met the inclusion criteria of the study and were considered in the analysis, representing 77% of those that initiated ART. Characteristics of these patients are outlined in Table 2. Study participants provided 9120 person-years of follow-up with an overall attrition rate of 39.22 per 100 person-years. At the end of the study period, 7491 (63%) study patients were alive and receiving ART from the clinics where they initiated ART. Of the remaining patients, 720 (6%) transferred to another clinic, 5 (<1%) suspended treatment, 1932 (16%) were lost to follow-up, and 1645 (14%) were known to have died. At 12 and 24 months, the proportion of patients retained in care was 68.1% (95% CI: 67.1% to 69.0%) and 58.7% (95% CI: 57.4% to 60.1%), respectively.
Table 3 illustrates the variability of patient volume and human resource levels of the clinics. The median number of monthly patient visits at the study clinics was 562 (IQR: 264-1141). The average number of clinicians per month providing clinical care was 2.89, and the mean number of pharmacy staff providing pharmacy services per month was 1.31. As a result, the median monthly number of clinical visits per clinician was 111 (IQR: 61-214) and the median monthly number of pharmacy visits per pharmacy staff was 359 (IQR: 142-609) among HIV care and treatment clinics.
Table 4 shows the hazard ratios (HRs) for the association of patient characteristics with attrition. Pre-ART CD4 count was more predictive of attrition in early follow-up than later follow-up. The relationship of patient volume, clinical staff burden, and pharmacy staff burden with attrition are illustrated in Table 5. Patients attending clinics with medium patient volume levels [HR = 1.45 (95% CI: 1.04 to 2.04)] and high patient volume levels [HR = 1.41 (95% CI: 1.04 to 1.92)] had a higher risk of attrition compared with patients attending clinics with low patient volume levels, but the test for trend was not significant (P = 0.198). Patients attending clinics with medium pharmacy staff burden [HR = 1.39 (95% CI: 1.07 to 1.80)] and high pharmacy staff burden [HR = 2.09 (95% CI: 1.50 to 2.91)] tended to have a higher risk of attrition than patients attending clinics with low pharmacy staff burden (P value for trend: <0.001). Patients attending clinics with higher clinical staff burden did not have a higher risk of attrition.
We also performed a sensitivity analysis. Findings were not dependent on categorization into tertiles and were similar to the complete-case analysis. After adjusting for year of treatment initiation, clinic experience and clinic location in exploratory analysis, the relationship between pharmacy staff burden and attrition became stronger. HR estimates did not significantly differ between early attrition and late attrition for patient volume or clinical staff burden. However, pharmacy staff burden had more attenuated risk estimates in early follow-up [medium pharmacy staff burden: HR = 1.66 (95% CI: 1.27 to 2.16), high pharmacy staff burden: HR = 2.42 (95% CI: 1.57 to 3.74)] compared with later follow-up [medium pharmacy staff burden: HR = 1.91 (95% CI: 1.21 to 3.01), high pharmacy staff burden: HR = 3.44 (95% CI: 1.95 to 6.08); P value for interaction = 0.092].
With data from 18 public sector HIV care and treatment clinics in Mozambique, we evaluated the impact of patient volume, clinical staff burden, and pharmacy staff burden on attrition from HIV treatment programs. No significant association was seen between clinical staffing burden and attrition, and although patients attending clinics with higher patient volumes did have an elevated risk of attrition, there was not a dose-response relationship. However, this study did show that patients attending clinics with higher pharmacy staff burden had a higher risk of attrition. This is an important finding as our results highlighted an additional potential area within the health system where interventions could be applied to improve the retention of these rapidly increasing patient populations in HIV treatment programs.
Strengths of the study included the standardization of protocols for care delivery, patient tracing, and data recording across the clinics included in the study. Additionally, we were able to assess associations between these characteristics and attrition with a large population of patients among clinics across different levels of the Mozambican health system. Furthermore, the databases had high levels of agreement with data from patient charts.
The principal limitation of our study was due to the observational nature of the research. Although we adjusted for measured patient characteristics to address concerns of confounding, the potential for unmeasured patient factors to bias our results existed. In addition, other unmeasured clinic-level factors correlated with our exposures could have driven the association. Another limitation was the sensitivity of the outcome. We used attrition due to our concern of differential outcome ascertainment; however, by doing so, we included an outcome, mortality, which was less sensitive to our clinic-level exposures of interest. The generalizability of these results to other settings should require careful consideration. Clinics were included in the study based on whether they utilized an electronic database and were not selected through a random process. The potential for exposure misclassification also existed. We did not have information regarding the number of working hours of clinicians and pharmacy staff. However, we would expect this misclassification to be nondifferential with respect to outcome status, which would attenuate our results toward the null on average.
This analysis built on our previous studies of the quality of HIV care in Mozambique with a larger number of health facilities.18,23,24 The clinics had substantial variability regarding the number of months offering treatment, number of adults initiating treatment, patient volume, and human resource levels. Disease severity metrics had larger hazard ratios during early follow-up but attenuated in later follow-up because they were measured at baseline. Education and year of treatment initiation had larger hazard ratios during later follow-up after the more clinically severe patients at enrollment either died or were lost to follow-up from the HIV treatment program.
With HIV care and treatment, the Mozambican health system implemented a chronic care intervention within facilities originally established for more acute or subacute care conditions. Attrition estimates of 32% and 41% in the HIV treatment program at 12 and 24 months after starting ART, respectively, was not optimal and was likely a result of patients starting ART late in their disease process and the presence of health system bottlenecks. Patients who received appointments for a particular day typically arrived to the clinic in the morning to begin their visit process. In this context, larger patient volumes could have stressed the general ability of the health system to cope, resulting in bottlenecked administrative systems and overwhelmed infrastructure. Our analysis of patient volume did not account for differences in facility infrastructure (ie number of outpatient consultation rooms) or human resources. This could explain in part why the trend test for patient volume was not statistically significant as clinics with larger patient volumes could have more space and resources to accommodate them.
Mozambique faced a critical shortage of pharmacy personnel with only 3 pharmacy staff per 100,000 people.12 In our analysis, pharmacy staff burden was an important predictor of attrition. Larger number of patients per pharmacy staff could result in longer wait times and shorter and lower quality patient-provider interactions. Monthly pharmaceutical visits to the clinic could be a substantial obstacle for patients to remain in treatment over the long term. A qualitative study of patients attending some of these clinics also identified pharmacy waiting times and transportation costs as barriers to ART retention.25 Research in other settings showed that length of wait and visit times and transportation costs influenced patient satisfaction and willingness to return for care.4,26,27 Clinical staff burden was not an important predictor of attrition which could be explained partly by the increase in available providers after task shifting of clinical responsibilities to nonphysician clinicians.23,28 In addition, fewer visits were recommended for clinical monitoring compared with returning monthly for pharmacy refills. A recent study by Bakanda et al29 in Uganda also found that clinician density did not affect attrition from ART programs.
The Mozambican Ministry of Health successfully scaled-up HIV care and treatment services in Manica and Sofala provinces of Mozambique. Challenges remain regarding the quality and efficiency of care delivery and retention in treatment. In conclusion, our study found a higher risk of attrition among patients attending clinics with larger numbers of patients per pharmacy staff. Our results highlight the key role that pharmacy staff play in providing HIV treatment services and argue for more initiatives to alleviate pharmacy staff burden in these settings. The encouraging preliminary findings of a community ART group model in Mozambique highlight an example of an out-of-clinic strategy which could reduce the frequency of patient visits to ease the impact of pharmacy staff burden on retention in treatment.30 Further understanding of the implementation process and the role health centers should continue to have in encouraging optimal use of medications and ensuring the health of patients is needed. If patients receive their refills outside of the clinic, other quality of care measures could potentially be compromised. Health systems strengthening measures, especially in the area of supply chain management, would need to compliment policy changes to more effectively and efficiently implement modifications. Additional research should identify the mechanisms by which pharmacy staff burden could result in attrition from HIV treatment and continue to identify potential interventions to alleviate the bottleneck.
We would like to thank the patients and providers at the clinics included in the study.
1. Ministério da Saúde. Maputo (MZ). Programa de tratemento antiretroviral (TARV)-dados TARV nacionais; 2009. Available at: http://www.misau.gov.mz/pt/hiv_sida/programa_de_tratamento_antiretroviral_tarv/dados_tarv_nacionais
. Accessed January 15, 2010.
2. UNAIDS. Epidemiological Fact Sheet on HIV and AIDS: Mozambique 2008 Update
. Geneva, Switzerland; UNAIDS. 2008. Available at: http://www.who.int/globalatlas/predefinedReports/EFS2008/full/EFS2008_MZ.pdf
. Accessed May 3, 2010.
3. Ministério da Saúde. National Plan for Health Human Resources Development (NPHHRD)
. Maputo, Mozambique
: Ministry of Health; 2008.
4. Wouters E, Heunis C, van Rensburg D, et al. Patient satisfaction with antiretroviral services at primary health-care facilities in the Free State, South Africa—a two-year study using four waves of cross-sectional data. BMC Health Serv Res
5. Stringer JS, Zulu I, Levy J, et al. Rapid scale-up of antiretroviral therapy
at primary care sites in Zambia: feasibility and early outcomes. JAMA
6. Habte D, Dussault G, Dovlo D. Challenges confronting the health workforce in sub-Saharan Africa. World Hosp Health Serv
7. Hosseinipour MC, Kazembe PN, Sanne IM, et al. Challenges in delivering antiretroviral treatment in resource poor countries. AIDS
. 2002;16(suppl 4):S177-S187.
8. Kober K, Van Damme W. Scaling up access to antiretroviral treatment in southern Africa: who will do the job? Lancet
9. Philips M, Zachariah R, Venis S. Task shifting for antiretroviral treatment delivery in sub-Saharan Africa: not a panacea. Lancet
10. Van Damme W, Kober K, Kegels G. Scaling-up antiretroviral treatment in Southern African countries with human resource shortage: how will health systems adapt? Soc Sci Med
11. Van Damme W, Kober K, Laga M. The real challenges for scaling up ART in sub-Saharan Africa. AIDS
12. World Health Organization. Working Together for Health: the World Health Report 2006
. Geneva, Switzerland: World Health Organization; 2006:209.
13. Institute of Medicine (US). Committee on Examining the Probable Consequences of Alternative Patterns of Widespread Antiretroviral Drug Use in Resource-Constrained Settings. In: Curran JW, ed. Scaling up Treatment for the Global AIDS Pandemic: Challenges and Opportunities
. Washington, DC: National Academies Press; 2005.
14. Chen L, Evans T, Anand S, et al. Human resources
for health: overcoming the crisis. Lancet
15. Narasimhan V, Brown H, Pablos-Mendez A, et al. Responding to the global human resources
16. United Nations Development Programme. Mozambique National Human Development Report 2007
. Maputo, Mozambique
: United Nations Development Programme; 2007:59.
17. Koepsell TD, Weiss NS. Epidemiologic Methods: Studying the Occurrence of Illness. Oxford, United Kingdom: Oxford University Press; 2003.
18. Sherr K, Micek MA, Gimbel SO, et al. Quality of HIV care provided by non-physician clinicians and physicians in Mozambique
: a retrospective cohort study. AIDS
. 2010;24(suppl 1):S59-S66.
19. Lin DY, Ying ZL. A simple nonparametric estimator of the bivariate survival function under univariate censoring. Biometrika
20. Lawn SD, Myer L, Harling G, et al. Determinants of mortality and nondeath losses from an antiretroviral treatment service in South Africa: implications for program evaluation. Clin Infect Dis
21. Van Buuren S, Boshuizen HC, Knook DL. Multiple imputation of missing blood pressure covariates in survival analysis. Stat Med
22. Rubin DB. Multiple Imputation for Non-Response in Surveys. New York, NY: John Wiley & Sons; 1987.
23. Micek MA, Gimbel-Sherr K, Baptista AJ, et al. Loss to follow-up of adults in public HIV care systems in central Mozambique
: identifying obstacles to treatment. J Acquir Immune Defic Syndr
24. Stubbs B, Micek M, Pfeiffer J, et al. Treatment partners and adherence to HAART in Central Mozambique
. AIDS Care
25. Lara J, Lambdin B. Exploratory analysis of the facility-level factors associated with ART abandonment in Central Mozambique
. Proceedings of the 137th Meeting of the American Public Health Association; 2009; Philadelphia, PA.
26. Camacho F, Anderson R, Safrit A, et al. The relationship between patient's perceived waiting time and office-based practice satisfaction. N C Med J
27. Hardon AP, Akurut D, Comoro C, et al. Hunger, waiting time and transport costs: time to confront challenges to ART adherence in Africa. AIDS Care
28. Hongoro C, McPake B. How to bridge the gap in human resources
for health. Lancet
29. Bakanda C, Birungi J, Mwesigwa R, et al. Density of healthcare providers and patient outcomes: evidence from a nationally Respresentative Multi-Site HIV Treatment Program in Uganda. PLos ONE
30. Decroo T, Telfer B, Biot M, et al. Distribution of antiretroviral treatment through self-forming groups of patients in Tete Province, Mozambique
. J Acquir Immune Defic Syndr