The studies were conducted in Mtwara, a rural region in southeastern Tanzania with more than a third of the population in the lowest national wealth quintile [28]. In 2012, malaria prevalence in Mtwara in patients of all ages seeking care for febrile illness was 15% among patients attending ADDOs and 32% among patients attending public health facilities [29,30]. Conversion of drug shops to ADDOs in Mtwara commenced in 2006, with all drug shops in Mtwara officially required to have upgraded to ADDO status by 2011. At the time of the study, some shops had not yet paid fees, received training, or been visited by inspectors but were tolerated as ‘prospective ADDOs’ (in this paper the term ADDOs is used to include both accredited outlets and ‘prospective ADDOs’). To support the increased availability of ACT in ADDOs, the government offered a one-day training that included treatment of malaria with ACT to dispensers with previous nursing training in Mtwara in August 2011.

In health facilities, a descriptive study was conducted to assess patient adherence to AL. The study in ADDOs was designed as part of a cluster-randomized trial to evaluate a text message intervention to improve dispenser knowledge of advice to provide to patients obtaining AL. Details of the intervention and results of the trial are presented separately [31]. Only data from the control arm, which did not receive any intervention, are presented here.

The target sample size for ADDO patients was based on the text message intervention trial [31], and a similar sample size was desired for public health facilities within the constraints of study resources. Based on adherence previously reported among patients obtaining ACT at public health facilities in Tanzania, which ranged from 75 to 98% [32-34], adherence among patients attending public health facilities was expected to be 75%, with lower adherence among patients attending ADDOs (60%). Assuming a hypothesized design effect of 2.5 and 20% loss to follow-up, 448 registered patients from 36 outlets in each sector (approximately 12 to 13 patients per outlet) would have 80% power to detect this difference with 95% confidence. Given the lower design effect observed in the study (1.5) and more interviewed patients than anticipated, the study was powered to detect a 10 percentage point difference between sectors, but was not able to detect if a smaller difference in completion rates between sectors was due to random error.

A list of all public dispensaries and health centres in Mtwara, excluding district hospitals, was compiled by visiting each district and interviewing the district medical officers. The health facilities were randomly ordered, and the first 40 health facilities were selected. Sampling of ADDOs was based on a census in Mtwara of all ADDOs, including prospective ADDOs, conducted prior to the text message intervention trial. From this list, ADDOs were selected sequentially at random, with all ADDOs within 400 m of a selected ADDO, or any ADDO where staff from a selected ADDO also worked, removed from the sampling frame. Forty ADDOs were randomized to the control arm [31].

From September through November 2012, dispensers at selected public health facilities and ADDOs were visited by study supervisors and given a standard introduction about study objectives. In order to limit patients’ awareness of the primary interest in adherence, which could have led to a biased assessment, dispensers were told the focus was how patients chose to treat fever and that some, but not all, patients would be visited at their homes. Dispensers were asked to fill out a registration form for all patients dispensed any treatment for fever, including day and time of the visit, the patient’s name, drugs dispensed, and a description of where the patient lived. Dispensers were provided with blister packs of AL to be dispensed in public health facilities to patients prescribed ACT, and in ADDOs to patients indicating an intention to purchase treatment for malaria. Study staff visited outlets every day to check and collect registration forms. While the intention had been to register 12 patients obtaining ACT for one week per outlet, the protocol was adjusted due to low attendance at some outlets to register all patients obtaining ACT for two to three weeks per outlet.

Eligible patients who obtained AL were identified from the registration forms and assigned patient identification numbers. Patients were visited at their homes three days later (day 4), and all attempts to locate and interview patients were recorded. Where written informed consent was given, patients or their caregivers were asked about demographic and socio-economic characteristics, treatment-seeking history, symptoms, and detailed information about each dose of AL taken. Patients were asked if dispensers provided each of several aspects of advice on AL (e.g., number of pills to take per dose, when to take second dose, etc.), and if so, what advice was given. Blister packs were also requested for a pill count. Since the P. falciparum histidine-rich protein II (HRP-2) persists in the blood following treatment, HRP-2-based malaria rapid diagnostic tests (mRDTs) (Pf-specific from ICT Diagnostics, Cape Town, South Africa) were conducted to indicate infection prior to treatment. Blood smears were collected to detect infection at the time of interview. Blood smears were stained in the field and transported to the Ifakara Health Institute, where they were double-read by two microscopists blinded to results from each other and the mRDT, with discrepant readings resolved by a third microscopist.

Adherence was defined in two ways [10]. Patients were considered to have verified completed treatment if they reported taking all doses by the time of the interview and, when available, a pill count verified that no pills remained in the blister pack. Where blister packs were not available, self-report alone determined if patients completed treatment. The second, more stringent definition included a time component, based on patient reports of the time each dose was taken using the Swahili times of day: alfajiri (early morning), asubuhi (morning), mchana (afternoon), jioni (evening), usiku (night), and usiku sana (late night). Patients were considered to have verified timely completion if they took the correct number of pills for each dose, and took the second dose at the Swahili time of day corresponding with eight hours after the first dose, followed by taking each of the remaining doses at the Swahili time of day corresponding with 12 hours after the previous dose, verified when possible by the absence of pills in the blister pack. The terms completed treatment and timely completion are hereafter used to refer to these definitions.

All patient and dispenser interview data were collected using personal digital assistants, and data extracted from study forms (census, registration and follow-up forms) were double entered into Microsoft Access databases. Data were analysed in Stata 11.0 (Stata Corporation, College Station, USA). Robust standard errors were used for percentages and 95% confidence intervals, with p-values reported for the Pearson design-based F test. Wealth quintiles describing socio-economic status were assigned to patients based on standard Demographic and Health Survey variables, using principal components analysis of the pooled sample of public health facility and ADDO patients [28]. Two analyses of factors associated with adherence were conducted: (i) a comparison of adherence between the public and private sectors controlling for patient characteristics; and, (ii) an analysis within each sector exploring the association of adherence with factors related to the care received at the outlet and patient characteristics.

In the analysis on the impact of sector on patient adherence, random effects logistic regression was used for both completed treatment and timely completion to compare the odds of adherence between private sector ADDO patients and public sector health facility patients, adjusting for patient characteristics identified a priori (age group, education and time between obtaining AL and interview) or those that made important changes to the odds ratio for sector in bivariate models. In this analysis, no adjustment was made for variables related to care received at the outlet (e.g., taking first dose at outlet, recalling correct instructions on how to take AL, etc.), as these factors might mediate the effect of sector (i.e., lie on the causal pathway between sector and adherence).

Within each sector, the association of variables related to care received at the outlet with 1) completed treatment, and, 2) timely completion was explored in univariate and multivariate models. Logistic regression with robust standard errors was used, as checks showed that the quadrature approximations for random effects models were not reliable. Patient characteristics and care-related variables that were associated with completed treatment or timely completion in either sector in unadjusted analyses were included in all four multivariate models.

All questionnaires, consent forms and other study documents were translated into Swahili and piloted prior to use. Written informed consent was collected from dispensers prior to census, patient registration, and interview and from patients or their caregivers prior to interview. The study protocol was approved by the ethical review boards of Ifakara Health Institute and London School of Hygiene and Tropical Medicine. CDC advisors provided technical assistance in design and analysis but were not engaged in data collection and did not have access to personal identifiers.

Data were collected from patients obtaining AL at all 40 selected health facilities and 37 of 40 selected ADDOs, with three ADDOs closed or refusing to participate. Of 604 registered health facility patients obtaining AL (median = 16 patients per outlet, range two to 32), 572 patients (95%) were interviewed. From ADDOs, 537 patients obtaining AL were registered (median = 17 patients per outlet, range one to 29), and 450 patients (84%) were interviewed. The most common reasons in both sectors for non-completion of interviews were not locating the patient’s home (38% at public facilities and 43% at ADDOs), or the patient having travelled out of the study region (28 and 16%).

Characteristics of patients differed between sectors (Table 1). ADDO patients were more likely to be male and older than those attending public health facilities. ADDO patients/caregivers were wealthier, with 32% in the least poor wealth quintile compared to 10% of public health facility patients, and were more likely to have finished primary school (72 vs 58%, p = 0.007). Over a third of patients in both sectors had previously sought care for their illness episode, many of whom had gone to a general store/kiosk or taken drugs stored at home or from a neighbour. Reported symptoms were similar across outlet type, except more patients from public health facilities had experienced respiratory symptoms (14 vs 7.5%, p = 0.007) and more ADDO patients had experienced body pain (30 vs 15%, p < 0.0001) or fatigue (18 vs 10%, p = 0.003), with few patients in either sector reporting convulsions or other signs of severe disease. ADDOs were much more likely than public health facilities to be located in urban areas, but outlets in both sectors were attended mostly by patients living within 2.5 km (about half-an-hour walk) from the outlet. ADDO patients were also more likely to have obtained AL later in the day, reflecting the fact that most public health facilities in Mtwara close for outpatient services by mid-afternoon, while ADDOs often stay open through the evening. Because of this, the day-4 interview was more likely to occur earlier (between 60–67 hours from the time the drug was obtained) for ADDO patients than for public health facility patients (25 vs 6%).Table 1

Patient characteristics by sector (%, 95% CI)

Table 2 compares care received at the outlet and patient status at interview for each sector. Half of the public health facility patients reported being tested for malaria, while 11% reported being tested in ADDOs, where diagnostic tests for malaria had not been officially introduced. Public health facility patients were slightly more likely to be told a diagnosis (64 vs 54%, p = 0.051) and more likely to take the first dose of AL immediately at the outlet (41 vs 10%, p < 0.0001), while ADDO patients were more likely to pay for AL (98 vs 29%, p < 0.0001). Similar percentages of patients using outlets in both sectors reported receiving advice on how to take AL from the dispenser, with 60% able to recall correct instructions on the number of pills per dose, number of doses per day, and number of days to take AL. Approximately 60% of patients in both sectors reported being told to take the second dose of AL after eight hours and to take each dose with food or milk. Public health facility patients were more likely to recall being told to finish all doses even if feeling better (78 vs 63%, p = 0.0006). However, less than 5% of patients treated in either sector reported being advised on possible side effects or on what to do in case of vomiting within half-an-hour of taking a dose.Table 2

Care received at outlet and patient status at interview by sector (%, 95% CI)

At the time of interview, approximately 14% of both public health facility and ADDO patients reported a current fever and 92% could play or work. More patients who had attended public health facilities tested positive by the mRDT performed by study staff during the interview (50% compared to 28% of ADDO patients, p = 0.001). However, by reference blood smear, indicating current infection status at the time of interview, only 2.9% of public health facility patients and 1.4% of ADDO patients were positive (p = 0.1).

Among public health facility patients, 74.5% (95% CI: 69.8, 78.8) completed treatment, compared with 69.8% (95% CI: 64.6, 74.5) among ADDO patients (p = 0.2). Timely completion was much lower and differed between sectors, with 46.0% (95% CI: 40.9, 51.2) of public health facility patients and 34.8% (95% CI: 30.1, 39.8) of ADDO patients taking the correct number of pills at the correct time of day for each dose (p = 0.003). Variables that made important differences to the odds ratio for sector in the bivariate models were wealth quintile, distance from home to outlet, and time of day AL was obtained, and these were included along with age group, patient/caregiver education, and time between obtaining AL and interview in the models comparing adherence between sectors. Patients in the two least poor wealth quintiles had higher adjusted odds of both measures of adherence compared to those in the poorest wealth quintile (Table 3). The adjusted odds of completed treatment for those who had finished primary school was 1.68 times that of patients who had not (95% CI: 1.20, 2.36; p = 0.003), but there was no evidence of an association with timely completion (aOR = 1.06, 95% CI: 0.77, 1.75; p = 0.9). Compared to patients who obtained AL in the morning, patients who obtained AL in the evening were similarly likely to complete treatment (aOR = 0.93, 95% CI: 0.50, 1.70; p = 0.8), but had much lower adjusted odds of timely completion (aOR = 0.35, 95% CI: 0.19, 0.64; p = 0.001). Furthermore, patients interviewed a longer time after obtaining AL (68–72 hours, 73–84 hours, and 85 hours or more) had higher adjusted odds of completed treatment and timely completion than patients visited 60–67 hours after obtaining AL (Table 3). When controlling for these patient characteristics, the adjusted odds of completed treatment for ADDO patients was 0.65 times the odds of completed treatment for public health facility patients (95% CI: 0.43, 1.00; p = 0.048), and the adjusted odds of timely completion for ADDO patients was 0.69 that of health facility patients (95% CI: 0.47, 1.01; p = 0.056).Table 3

Effect of sector on adherence controlling for potential confounders ¹

Unadjusted associations of patient characteristics and care received at the outlet with completed treatment and timely completion are presented for each sector in Additional files 1 and 2. Variables not associated with either adherence measure in either sector, such as patient sex, having sought previous treatment, report of specific symptoms, and paying for AL, were not included in the sector specific multivariate analyses shown in Table 4. In contrast with the analyses comparing the effect of sector on adherence, in these sector specific models there were no clear patterns of association between education and socioeconomic status and either measure of adherence. Similar to the previous analysis for the effect of sector, patients obtaining AL in the afternoon or in the evening had lower adjusted odds of timely completion than those obtaining AL in the morning, although this was the case only in public health facilities. In both public health facilities and ADDOs, longer time between obtaining AL and interview were again strongly associated with completed treatment. However, there was no evidence of an association with timely completion in public health facilities, and in ADDOs only being interviewed 72–84 hours after obtaining AL (not 68–72 hours or 85 hours or more) compared to 60–67 hours was associated with timely completion. In addition, public health facility patients, but not ADDO patients, sleeping under a bed net the night before the interview, having experienced fever symptoms, and living within 2.5 km of the outlet was associated with completed treatment, while seeking care within two days of fever onset was associated with timely completion.Table 4

Multivariate analyses of factors associated with adherence, by sector

Factors related to care received at the outlet varied by sector in their associations with both adherence measures (Table 4). In the public sector, reporting being tested for malaria at the outlet was not associated with completed treatment (aOR = 1.30, 95% CI: 0.82, 2.04; p = 0.3), but there was weak evidence of an association with timely completion (aOR = 1.47, 95% CI: 0.96-2.29; p = 0.078). Among ADDO patients, however, the adjusted odds of completed treatment and timely completion were lower for those who reported being tested compared to those who did not report being tested (aOR = 0.39, 95% CI: 0.17-0.85; p = 0.018 and aOR = 0.48, 95% CI: 0.24, 0.97; p = 0.041). There were no evident associations of taking the first dose of AL at the outlet with completed treatment in either sector or timely completion in ADDO patients, but the adjusted odds of timely completion were higher among health facility patients who took the first dose of AL at the outlet (aOR: 2.11, 95% CI: 1.46, 3.04; p < 0.001). Recalling correct instructions given by the dispenser on the AL regimen was strongly associated with both measures of adherence in public health facility and ADDO patients. In public health facilities, reporting that the dispenser used the packaging as a visual aid to explain how to take AL was also associated with timely completion (aOR = 1.77, 95% CI: 1.12, 2.80; p = 0.022). In addition, the adjusted odds of timely completion among ADDO patients who recalled being told to take the second dose after eight hours were 1.77 times that of their counterparts (95% CI: 1.12, 2.80; p = 0.015), but recalling this advice was not associated with either measure of adherence in public health facility patients. Reporting being told to complete all doses of AL even if feeling better was also not associated with either adherence measure, except for a lower adjusted odds of timely completion in public health facility patients (aOR = 0.44, 95% CI: 0.28, 0.70; p = 0.001).