A nationally representative outlet survey was conducted in 2010 as part of the Independent Evaluation of the Affordable Medicines Facility–malaria (AMFm), a multi-country antimalarial subsidy program.12 The outlet survey covered public, private not-for-profit, and private for-profit antimalarial outlets, but we focus here on the private for-profit sector (pharmacies, DLDB, ADDO, and general retailers) where ACA samples were collected. Data collection took place between September and November 2010 before AMFm was implemented. In total, 48 wards throughout mainland Tanzania's 21 regions‡ ‡

Tanzania had 21 regions at the time of the study, though this has subsequently been increased to 25. Results are presented based on the 21 regions at the time of data collection.

The sample size was determined based on the needs of the independent evaluation (full details available in Tougher and others12), which required 305 outlets stocking antimalarials in both urban and rural domains. Estimates of the average numbers of outlets per ward were used to estimate the number of wards required to reach this target, leading to the selection of 9 urban and 39 rural wards.

After verbal informed consent, full interviews were conducted in outlets with an antimalarial in stock at the time of visit or within the previous 3 months, with data collected using personal digital assistants (PDAs). Questions about outlet characteristics were asked using a structured questionnaire, and details about every antimalarial in stock at the time of visit were recorded.

A sample of every ACA present at the time of visit was purchased from every private for-profit outlet that reported stocking any ACAs. For prepackaged ACAs, one packet for each age group was purchased; for ampoules/vials, suppositories, and packets of oral pediatric granules for suspension, at least six units were purchased; and for bottles of syrups and powders for suspensions one bottle was purchased.

Each drug sample was sealed in a waterproof bag with desiccant at the time of purchase. All samples were boxed and stored at ambient temperature for maximum 3 months before being sent to the London School of Hygiene & Tropical Medicine (LSHTM; United Kingdom) by air. At LSHTM, samples were stored in the dark at ambient temperatures (22°C) until and during analysis.

We measured the quantity of API present for all artemisinin derivatives. In addition, we measured the API amount for partner drugs in selected ACTs. Because of capacity constraints, it was necessary to limit the partner analysis to the most common partner drugs for each generic artemisinin ingredient: lumefantrine, piperaquine, and mefloquine (used in the ACTs Alu, artemisinin–piperaquine, dihydroartemisinin–piperaquine, artesunate–amodiaquine and artesunate–mefloquine).

The physical dimensions and weight of each antimalarial tablet were recorded prior to processing for analysis, to fully document sample characteristics before preparation. Formulations were analyzed for the amount of each API present using high-performance liquid chromatography with ultraviolet photo-diode array detection (HPLC-UV-PDA). Each tablet was pulverized, and dissolved in solvent depending on the stated API: artesunate, artemether, dihydroartemisinin, amodiaquine, sulfamethoxypyrazine/pyrimethamine, and SP were dissolved in methanol; samples containing lumefantrine (LUM) were dissolved in 10% acetic acid in methanol, mefloquine (MF) in methanol/2.0 N hydrochloric acid (MeOH/2.0 N HCl; v/v) and piperaquine (PIP) in methanol/0.1 M HCl (1:1; v/v). Solvent extracts were sonicated followed by centrifuging, and the supernatant injected into the HPLC column to determine the amount of API present. Syrups, powders, suspensions and suppositories were dissolved in methanol. Injectables where the stated API carrier was coconut oil were also dissolved in methanol, with injectables where the carrier was peanut oil dissolved in isopropanol: hexane (v/v; 16:1).

All samples analyzed at LSHTM were sent to the U.S. Centers for Disease Control and Prevention Laboratories, Atlanta, GA, and the Georgia Institute of Technology, Atlanta, GA, for HPLC confirmatory analysis and mass spectrometry (MS) screening, respectively (both groups were blinded to the LSHTM results). Additional laboratory analysis details are given in Supplemental Information.

Data analysis was conducted in Stata version 12 (StataCorp LP, College Station, TX). The primary outcome was “drug quality,” defined as the percentage of the stated amount of API detected in each drug component. Under the World Health Organization (WHO) International Pharmacopoeia, the acceptable range for artemisinin-based antimalarials and their companion drugs is to contain 90.0–110.0% active ingredient stated on the label. Reflecting the typically observed experimental variation in HPLC laboratory testing where less than 10 samples of each ACA formulation were tested, we adopted a wider tolerance band of 85–115% for the purposes of this study. Thus, samples were considered to be of “acceptable quality” in terms of the artemisinin derivative component if these samples contained 85–115% artemisinin derivative API as labeled, and acceptable quality in terms of both the artemisinin derivative and partner drug components if they contained 85–115% API, as labeled for both components.

Results were weighted to reflect the sampling strategy for the outlet survey.35 The analysis was corrected for clustering at the district level and urban/rural stratification using Stata SVY commands. Statistical tests were evaluated at P ≤ 0.05 level of precision.

The outlet survey included data on several potential risk factors for poor quality medicines, including characteristics of the products and the outlets from which they were purchased. Outlet characteristics included type of outlet (pharmacy, drug store [DLDB or ADDO], general store), and urban or rural location. Product characteristics included generic components, date of manufacture, date of expiry, stated manufacturer, stated country of manufacture, dose form (e.g., suspension, tablet, injectable), whether the drug was WHO prequalified at the time of the survey, whether the drug was a fixed-dose combination, and retail price. “Cheap” drugs were defined as drugs with a price in the lowest quartile of all ACAs sampled, stratified by dosage form. Prices were calculated using adult equivalent treatment doses (AETDs), the amount of a drug needed to completely treat a 60-kg adult.31 For example, to calculate the price per AETD of a pediatric package of Alu with six standard tablets (20 mg artemether and 120 mg lumefantrine), the price would be multiplied by four to calculate the cost for an adult equivalent dose of 24 tablets.

The association between poor ACA quality and outlet and product characteristics was examined using regression analysis, taking survey design into account. Unordered categorical variables were tested using the Wald test to estimate the overall statistical significance of categorical variables, for example, dose form. Sensitivity analysis was undertaken using a narrower API range for acceptable quality of 90–110% stated API.

Ethical approval was obtained from the Tanzanian National Institute for Medical Research, the Institutional Review Board of the Ifakara Health Institute, the London School of Hygiene & Tropical Medicine Research Ethics Committee, and from the Institutional Review Board of ICF International.

In total, 3,029 private for-profit retail outlets were visited, of which 512 had an antimalarial in stock on the day of visit (Table 1). From the outlets stocking ACAs, 1,737 ACAs were purchased and analyzed for API content: 55 ACAs were documented in outlets but not analyzed because either it was not possible to purchase the ACA or the purchased ACA could not be linked to the relevant outlet at the time of analysis. By generic class, the most frequently analyzed ACA was artemether: 604 doses were analyzed, of which 504 were stated to be Alu, the first-line antimalarial (Table 2). In all, samples from 177 different stated batches were analyzed.

Of all outlets with any antimalarials in stock at the time of the visit, nearly all pharmacies reported having at least one member of staff with a health-related qualification (97.6%), compared with only 7.0% general retailers (Table 3). Of all outlets with antimalarials in stock, 21.6% had an ACA, and ACA-stocking outlets were identified in 16 of the 21 regions. Stocking patterns varied substantially across outlet types, with 96.0% pharmacies stocking an ACA, compared with 24.5% drug stores (DLDB or ADDO) and 4.6% general retailers. A WHO prequalified ACT was stocked by 71.0% pharmacies, 11.9% drug stores, and 4.6% general retailers. No drug stores or general retailers stocked artemisinin monotherapies, but 39.8% pharmacies stocked injectable artemisinin monotherapies and 2.1% pharmacies stocked oral artemisinin monotherapy.

Overall, 58.1% ACAs were stocked by drug stores, and 67.7% were from urban areas (Table 4). The majority of products were combination therapies (97.0%), and were in tablet form (86.1%). The most commonly stated regions of manufacture were Asia (45.3%), Africa (27.6%), and Europe (20.7%). Only a quarter of ACAs were WHO prequalified (25.7%). The largest proportions of non-prequalified ACAs were stated to be manufactured in Asia (43.9%) and Africa (36.7%), while a smaller proportion were stated to be manufactured in Europe (18.9%).

Only 2.0% ACAs had expired at the time of purchase. However, 67.4% had expired by the time of analysis because of delays in processing and analyzing the samples. Of these, the majority had expired within 12 months of analysis (43.5%), but a substantial minority had expired more than 1 year before the time of analysis (23.5%). The implications of these delays are discussed below.

All drugs analyzed contained the stated artemisinin derivative. Overall 95.9% samples contained between 85% and 115% recommended API for the artemisinin component, with 1.6% containing below 85% and 2.4% containing above 115% (Figure 1). By type of artemisinin derivative, 97.9% artemether and 93.9% artesunate-based drugs were of acceptable quality by this definition. Dihydroartemisinin-based drugs had the highest percentage of products with below 85% stated artemisinin API, while artesunate-containing drugs had the highest percentage of products with more than 115% stated artemisinin API.

The MS analysis conducted at the Georgia Institute of Technology confirmed that all samples contained the stated artemisinin and partner active ingredient, and no other active ingredients were identified. There was a reasonable level of agreement between the HPLC analyses conducted at LSHTM and the U.S. Centers for Disease Control and Prevention (Supplemental Information).

Figure 2A shows the proportion of poor quality ACAs by region in Tanzania based on the artemisinin component. Poor quality ACAs were purchased from 10 of the 16 regions where ACAs were found. In regions where poor quality ACAs were found, the prevalence ranged from below 1% to 9%.

Table 5 shows the percentage of poor quality drugs based on the artemisinin component by risk factor. On the basis of this bivariate analysis, four risk factors were associated with poor quality at P < 0.05: not being WHO prequalified, being relatively expensive (not in lowest price quartile), dose form, and type of artemisinin derivative. Only 0.5% WHO prequalified drugs were of poor quality, compared with 5.4% of those not prequalified. Nearly one-third (29.6%) of pediatric granules were of poor quality, compared with 3.4%, 5.4%, and 3.0% for tablets, suspensions, and injectables, respectively. There was less variation in poor quality by type of artemisinin derivatives: 6.1% artesunate derivatives were of poor quality compared with 4.7%, 2.5%, and 2.1% for dihydroartemisinin, artemisinin, and artemether, respectively. Type of outlet, urban/rural location, whether the product was an ACT rather than a monotherapy, and stated region of manufacture were not found to be associated with poor quality. Expiry at date of analysis was of borderline significance.

Data were too sparse for robust multivariable analysis. However, given the possible association of expiry by date of analysis with quality, we investigated the relationship between drug quality and other risk factors by recalculating their odds ratios (ORs) adjusting for expiry by date of analysis. This adjusted analysis was performed for all other risk factors where P < 0.2 in the bivariate analysis (Supplemental Table 1). Adjustment led to no major change in ORs, with the exception of dihydroartemisinin where the OR decreased from 2.3 to 1.9. Only WHO prequalification remained significantly associated with quality after this adjustment (OR 0.1, 95% confidence interval [CI]: 0.01–0.34, P = 0.002).

Laboratory analysis was also conducted on three partner drugs (lumefantrine, piperaquine, and mefloquine) used in the ACTs Alu, artemisinin–piperaquine, dihydroartemisinin–piperaquine, and artesunate–mefloquine. Considering both artemisinin and partner drug quality, 87.9% ACTs were of acceptable quality (Table 6) with 5.1% partner drugs containing higher than the 115% API limit and 3.7% below the 85% API limit. A small proportion of drugs had components that both fell either above (1.6%) or below (0.3%) the 85–115% API range. Poor quality ACTs based on quality of either component were purchased from 10 of the 16 regions where ACTs were found, with prevalence ranging from 1.4% to 26.5%, across regions (Figure 2B).

Table 7 shows the proportion of ACTs that were of poor quality based on both components by risk factor. As with the analysis of artemisinin derivative components alone, not being WHO prequalified, dose form, and type of artemisinin derivative were associated with poor quality based on both components. In addition, expiry at date of analysis was also a risk factor for poor quality when both components were considered. Around one-fifth of artesunate (19.2%)- and dihydroartemisinin (21.1%)-based ACTs were of poor quality. This was nearly double the proportion of poor quality artemisinin-based ACTs (11.6%), and almost seven times that for artemether-based ACTs (3.1%). Price was of borderline significance.

Adjusting for expiry at date of analysis led to no major changes in ORs, with the exception of Europe as stated country of manufacture where the OR decreased from 3.0 to 0.9 (Supplemental Table 2). The only risk factors remaining significant at 0.05 level after adjustment were WHO prequalification (OR 0.04, 95% CI: 0.01–0.2, P < 0.001) and type of artemisinin derivative (OR ranged from 5.2 to 7.3, P = 0.01).

To determine how sensitive results were to the percent API cutoffs used to define poor quality, the analysis was repeated using a narrower range of 90–110%. Using these tighter cutoffs, the percent of products defined as poor quality on the basis of the artemisinin component would have risen from 4.1% to 17.6% and on the basis of both components in ACTs from 12.1% to 30.6% (Figure 3).