Adherence to prescribed artemisinin-based combination therapy in Garissa and Bunyala districts, Kenya

The study was undertaken in two districts: Bunyala district, Western Province, close to Lake Victoria with perennial malaria transmission and community-based parasite prevalence among children in excess of 40% [29], and Garissa district, North-Eastern Province, a semi-arid region with acutely seasonal malaria transmission and childhood parasite prevalence of less than 1% during the dry season [29]. Two government health facilities were selected in Bunyala district (Budalangi dispensary and Mukhobola health centre), and three dispensaries in Garissa district (G. K. Prison, Utawala and Young Muslim). Mukhobola had an outpatient, as well as an inpatient facility, whilst the remaining four had outpatient facilities only. At outpatient facilities, the clinical officers or community health nurses were responsible for prescribing and dispensing AL.

Patients were recruited from selected health facilities between September and December 2009 in Bunyala, and between July and August 2010 in Garissa. Inclusion criteria were: testing positive for malaria parasites using slide microscopy or rapid diagnostic tests, having a weight of ≥ 5 kg and being more than 5 months of age. If these patients had reported no other anti-malarial use during the preceding two weeks, had no signs of complicated malaria or other severe disease, and did not have another household member participating in the study, they were recruited until 646 patients had been enrolled in Budalangi in three age groups (< 5 years, 5-14 years and ≥ 15 years), and 272 had been enrolled in Garissa in two age groups (< 14 years and ≥ 15 years). Given the low transmission of malaria during the dry season in Garissa, the sample size had been reduced to accommodate for fewer patients attending the health facilities. Correspondingly, sample size was based on a projected adherence of 70% in Garissa and 60% in Budalangi, 10% precision in the point estimate, a 5% type-1 error and an estimated 20% loss to follow-up (EpiInfo 3.5.1, CDC, WHO).

All patients that fulfilled the entry criteria and were prescribed AL were asked to keep their blister packs once they had completed the medication, and details were recorded of each patient's residential address and clinic card. Patients were not told why their details were being taken or why they were required to keep their blister packs, thus ensuring patient adherence was not affected by knowledge of the survey.

On the fourth day after recruitment, patients were followed-up at home, by which time all doses should have been taken. All respondents, in this case patients aged 15 years or more and caregivers of patients aged less than 15 years, had the study explained to them in a language they were most familiar with, and written informed consent was obtained. Interviews were conducted by field officers using a structured questionnaire written in English and translated into KiSomali (Garissa district), BaLuya (Bunyala district) or KiSwahili where appropriate (see Additional File 1). The questionnaire included information on socio-demographic characteristics of the patient/caregiver and of their household (household size, education level, employment status), details of information provided by the prescribers at the health facilities (whether or not they were told how to take AL correctly), respondent knowledge of the AL dosing regimen and the patient's acceptance of the AL treatment (whether or not any dislikes or side-effects were reported). Respondent knowledge regarding the ACT dosing regimen was assessed by asking respondents directly how AL should be taken, and the number of correct statements given was totalled. Correct statements included taking the first dose immediately, taking the second dose after 8 hours, taking the remaining doses in the morning and evening, taking doses with milk/fatty foods, and taking all the tablets. Additionally, blister packs were directly observed and the patient's adherence status was then recorded. Patients identified on the day of follow-up who had not completed the course were asked why, and were then advised to return to a health facility for follow up treatment.

Open-ended questions were reviewed and grouped into thematic categories that were coded for data entry. Results were double entered into EpiInfo 3.5.1 (CDC, WHO) and all data were analysed in STATA 11.0 (Stata Corporation, College Station, Texas). Patients who had a blister pack still containing ACT when interviewed were considered 'definitely non-adherent', and patients were categorized as 'probably non-adherent' if they or their caregivers claimed the doses were not completed but did not show a blister pack. 'Probably adherent' patients were those who claimed to have completed all the doses and could show and empty blister pack, as well as patients who could not show a blister pack, but claimed to have taken all the tablets.

To identify predictors related to adherence, data from the two study sites were combined and chi-square tests were first performed in a univariate logistic regression model. Outcomes were grouped by those considered 'probably adherent' and those considered either 'probably' or 'definitely non-adherent'. Factors tested at univariate logistic regression included demographic information relating to the respondent, patient and head of household, as well as health-seeking behaviour and patient acceptance of AL. To increase statistical power and simplify interpretations, multilevel categorical variables were collapsed into binary ones. A forward stepwise selection procedure was performed. Predictors found to be significant at the P ≤ 0.25 level were entered into a multivariate logistic regression model with remaining non-significant predictors introduced one at a time to detect for additional confounders. These remained in the model if the OR of other predictors in the model changed by greater than 20%, though no additional confounders were found. A backwards stepwise elimination procedure was performed with a P ≤ 0.05 criterion for remaining in the model. All regression analyses were undertaken adjusting for clustering at health facility level. Any questionnaires with missing data were not included in the regression models.

Ethical approval for the study was provided by the Kenya Medical Research Institute/National Ethical Review Committee (protocol number 1742).

A total of 893 patients in Bunyala district were enrolled; 87 (9.7%) were excluded from interview because members of their household had already been interviewed, 80 (9.0%) refused participation and 80 (9.0%) were lost to subsequent follow-up. Of the remaining 646 patients, approximately equivalent recruitment was obtained across the three age groups and gender. A total of 351 patients were enrolled in Garissa. Of these, 43 (12.3%) did not meet the inclusion criteria as a member of their household had already been interviewed, two (0.6%) refused to participate and 34 (9.7%) were lost to follow-up and thus were not included in the final analysis. Approximately equivalent age group recruitment was observed in the final sample of 272 patients, though these groups were significantly different regarding sex distribution. The characteristics of the recruited study populations in both districts are shown in Table 1.

Of the 918 patients treated with AL, 697 (76.1%) retained and showed a treatment blister pack during the home visit. Overall, 588 (64.1%) patients claimed to have taken all tablets and were considered 'probably adherent', of whom 69.2% (407/588) showed empty blister packs during the home visit. A further 291 (31.7%) patients had AL doses remaining in their blister pack and were therefore considered 'definitely non-adherent'. Only 39 (4.3%) patients reported they had taken their drugs incorrectly and had not retained their blister packs. These patients were therefore categorized as 'probably non-adherent'. Nearly equal proportions of adherence were observed in the two study sites (Table 2).

Univariate logistic regression analysis compared 19 predictors of patients who were classified as adherent (n = 588) versus those that were not adherent (n = 330). Fourteen predictors with P-value ≤ 0.25 were fitted in a multivariate logistic regression model: patient age (< 15 and ≥ 15 years); respondent and head of household age (< 25, 25-50 or > 50 years); respondent education level (never attended, primary level or secondary level and above); if the respondent and head of household was employed; household size (0-5 and ≥ 6); if patient had waited more than 24 hours to seek treatment; if respondent had taken AL before; if respondent had seen AL before, if patient had any dislikes or side-effects to the treatment, patient knowledge (no knowledge of how to take AL and at least one correct statement) and if the patient had been given any instructions as to how to take AL at the health facility. Six variables were significantly (P < 0.05) associated with the outcome after allowing for potential confounders (Table 3). The strongest predictor of adherence concerned patient knowledge of the ACT regimen; if a patient was able to cite at least one correct instruction as to how to take AL, they had 1.76 (95% CI = 1.32-2.35) the odds of being fully adherent than a respondent () that could cite none. Other significant factors included whether a respondent was aged between 25-50 years as opposed to less than 25 years (odds ratio (OR) = 1.65, 95% CI = 1.10-2.48); whether a patient was aged 15 years or more as opposed to less than 15 (OR = 1.37, 95% CI = 1.02-1.85); whether a respondent had seen the drug before or not (OR = 1.46, 95% CI = 1.08-1.98); whether a patient had reported dislikes to medication or not (OR = 0.62, 95% CI = 0.47-0.82) and if a respondent had waited more than 24 hours to seek treatment as opposed to seeking treatment in ≤ 24 hours after becoming symptomatic (OR = 0.73, 95% CI = 0.54-0.99).