Evaluating the Impact of a Comprehensive School Health Program in Zambia

Since 2015, our partner NGO Healthy Learners has partnered with the Zambian Government to develop a comprehensive and scalable school health program (SHP), making schools an entry point into the healthcare system. The SHP model consists of 4 key features: 1) building capacity by training teachers to become community health workers and building and equipping a school health room; 2) diagnosing and treating sick learners in school health rooms with the help of the ThinkMD clinical decision support system, or referring learners to the local health centre with a fast-track form; 3) proactive monitoring of absent and sick learners (e.g., following up after health centre referrals, a "buddy" system to check up on absent learners); 4) prevention through school health education (delivered by trained teachers) and supply of preventive care in partnership with local health facilities (e.g., vitamin A, deworming, etc.)

2024-09-02

2026-11-27

1. Any healthcare utilisation
2. Healthcare utilisation conditional on need
3. Attendance rate

1. Any healthcare utilisation
The proportion of learners who sought formal care (any government health centre or hospital, or the school health room). Equals one if reported at least once during the health diary observation period.
[Time Frame: 8-week period across term 1 and term 2 2026]

2. Healthcare utilisation conditional on need
Conditional measure of healthcare use, restricted to the sample of children who have experienced at least one ‘serious’ illness episode, defined as one which requires medical attention according to WHO c-IMCI and IMAI guidelines. The outcome is the ratio of the number of illness episodes during which a child seeks medical care over the total number of serious illness episodes experienced by a child, for all children who have experienced at least one serious illness episode over the period.
[Time Frame: Eight week diary observation period]

3. Attendance rate
We assess the program’s effect on school attendance using data collected through repeated, unannounced spot checks at each school. Our primary attendance measure will be based on the sample of children enrolled in the study at baseline and confirmed to still be enrolled in their original school at the time of the visit. In this sample, the attendance rate, measured at each attendance spot check, will be the proportion of those children confirmed present in class on the day of the visit.
[Time Frame: 24 months after intervention start]

4. Malaria prevalence
5. Moderate-to-severe anaemia
6. Underweight
7. Diarrhoea prevalence
8. Haematuria prevalence
9. Menstrual health knowledge
10. Menstrual health management
11. Menstrual health stigma
12. Student learning (numeracy, literacy, science)
13. Students' working memory

School attendance (complementary measures):
14. Attendance + retention (based on panel of learners)
15. Attendance with tracking (based on panel)
16. Register-based attendance
17. Illness-related absence (captured in health diaries)

4. Malaria prevalence: Proportion of children testing positive for malaria
Testing for the presence of P. falciparum. The tests are performed using a rapid diagnostic test (RDT) on the sample of children taking part in the health diaries (testing will occur at the end of the period).
[Time Frame: End of health diaries in term 1 and term 2 2026]

5. Proportion of children with moderate or severe anaemia
Haemoglobin concentration measured using a HemoCue photometer in the sample of children taking part in the health diaries (testing will occur at the end of the period). Anemia severity categories will be based on the WHO classification which depends on the age and sex of the child.
[Time Frame: Half of the sample captured in term 1 2026 and half in term 2 2026]

6. Proportion of children who are underweight
Using height and weight measured for all learners surveyed at endline, we will construct weight-for-age z-scores and determine the proportion of children who are underweight.
[Time Frame: Endline]

7. Proportion of children with diarrhoea
Incidence of diarrhoea - carer report of the child experiencing three or more loose or watery stools.
[Time Frame: Over the eight week follow-up period across term 1 and 2 in 2026]

8. Haematuria prevalence
An indicator for incidence of haematuria (blood in urine) reported by the caregiver.
[Time Frame: in the last 7 days, household survey]

9, 10 and 11: Menstrual health knowledge, management, and stigma
We examine the impact of the programme across three domains: knowledge, practices, and stigma. For each domain, we draw on questions developed and validated in prior studies (Austrian, Kangwana et al. 2021, Kansiime, Hytti et al. 2022, Hennegan, Hasan et al. 2024, Macours, Vera et al. 2024). For each sub-scale, we will construct a summary index using principal component analysis, standardized as a z-score using the mean and standard deviation of the control group for ease of interpretation of the results. In analysis, we will look at the impact of the programme on each of these domains separately, as well as combined.
[Time Frame: Endline]

12. Student learning (numeracy, literacy, science)
Learning is measured using individually administered numeracy, literacy, and science examinations. Items for these exams are aligned with the Zambian Ministry of Education national curriculum for Grades 1 through 7. Each assessment will be conducted one-on-one by trained enumerators at the start of the learners’ interviews. Items are developed by local curriculum experts and calibrated through pilot testing to ensure grade-appropriate difficulty. We will also borrow items from international test banks (e.g., TIMSS, EGRA, EGMA). Finally, we will classify items using Bloom’s taxonomy to test for differences in effects across higher- and lower-order thinking skills. In analysis, we will look at each domain separately as well as a combined score.
[Time Frame: Endline]

13. Students' working memory
We measure students’ working memory using the Digit Span task, a widely used short-term memory and working memory assessment. Enumerators orally present a sequence of digits at a rate of one per second, and students are asked to repeat the digits in the same order (Digit Span Forward) and then in reverse order (Digit Span Backward). The task measures students’ auditory attention, concentration, and working memory capacity, which are foundational components of general cognitive functioning and predictive of academic achievement. Scores are recorded as the total number of correct sequences reproduced, following standard administration and scoring procedures.
[Time Frame: Endline]

School attendance (complementary measures):

14. Combined measure of attendance and retention
Proportion of children confirmed present in class on the day of the visit, out of the total number of children enrolled at the school at baseline and still alive. In this measure, we effectively treat children who may have transferred to another study school as ‘absent’, dropped out, or temporarily relocated.
[Time Frame: Five random spot checks across 18 months]

15. Attendance rate including learners transfered to other study schools
Proportion of children confirmed present in class on the day of the visit either in their original school or in another school to which they have transferred if that school is one of the study schools, out of the total number of children enrolled at baseline and still alive. Following an Intention-to-Treat principle, we will analyse participants based on the arm they were originally assigned to, even if they moved to a different study arm school.
[Time Frame: Five unannounced spot checks across 18 months]

16. Cross-section attendance
Additional measure of attendance, focusing on students in grades 1, 3, 5, and 7 at the time of the spot check. This register-based measure is defined as the proportion of children on the register who are present on the day of the spot check.
[Time Frame: Four randomised spot checks conducted across 12 months]

17. Illness- or menstruation-related absence
From the diary's daily records, caregivers flag whether the child attended school each day and records the reason for each absent day. The reasons are: illness (1), was menstruating (2), farming (3), school closure or school event (4), bad weather (5), religious reasons (6), market day (7), public holiday (8), fishing (9), mining (10), working with parents (11), no uniform/shoes/books (12), and other (-97). We treat a day as an illness or menstruation-related absence when the reason is (1) or (2). We restrict the sample to school days by dropping weekends and public holidays. Our measure is the share of the child's school days in the diary window lost to health-related absence: health-related absence days divided by total school days observed.

We will assess the impact of the SHP in a parallel-arm, cluster-randomised controlled trial (cRCT) in 225 schools (clusters), with a 1:1.25:1.5 allocation ratio of clusters to delivery of (i) the usual school health activities (the control group – 75 schools), (ii) the SHP intervention (90 schools), and (iii) mass drug administration. The trial will be conducted in primary schools (including children from Grade 1 to Grade 7, aged 7-14 years old) in six districts in the Copperbelt (Luanshya, Chingola, Masaiti) and Luapula (Samfya, Mwense, Kawambwa) provinces, covering a range of urban, peri-urban, and rural areas, as well as varying levels of endemicity of infectious diseases (malaria and worms). Of the 286 government schools currently in these districts, Healthy Learners identified 250 that meet their eligibility criteria for programme implementation (sufficient staffing and reachable during the rainy season). Of those, we randomly selected 225 for the trial. To ensure a balanced sample with respect to geography and school environment, the randomization was stratified by district and school size.

In each school, we will collect data from (i) a random sample of 60 learners in Grade 1 (G1), G3 and G5 in 2024 who are present at the time of the baseline survey, whose parent/guardian will provide consent and who will assent to the study; (ii) a random sub-sample of 30 from sample (i) invited to take part in the symptom diary sub-study; (iii) a yearly sample of four classes (G1, G3, G5 and G7) comprising about 120-160 students in total, for the attendance study; (iv) a sample of one headteacher and up to 10 teachers present at baseline and who consent to the study; (v) a sample of 145 healthcare facilities serving the study schools and up to three health workers who staff these facilities.

Not available

Randomisation algorithm in STATA

School

Yes

225 schools

13,300 households; 13,300 learners; 225 school administrators; 1,700 teachers; 145 health facility administrators; 300 health workers

90 schools in SHP arm; 75 schools in standard of care control group; 60 schools in mass drug administration arm

We computed power calculations and concluded that our study was well-powered for the two co-primary outcomes defined above. We use a significance level of 5% and estimate minimum detectable effect (MDE) sizes with 80% power. We report MDEs based on intention-to-treat effects. Unconditional use of health care. We expect the most direct effect of the school health programme to be an increase in the healthcare utilisation (HA1). In a pilot study carried out in October 2025, 31% of children surveyed sought care at least one. Assuming an intra-cluster correlation of 0.30 (higher than the one estimated during the pilot), we are powered to detect an increase of 10.5 percentage point in the probability of ever seeking care in the SHP arm relative to the control arm. Conditional use of needed health care. We expect the biggest and most direct effect of the school health programme to be an increase in the utilisation of health care when needed (HA1). This outcome is conditional on a child experiencing at least one serious illness episode over the observation period. Based on results obtained from a pilot study carried out in October 2025, we expect that about 50% (i.e., approximately 12) of the 24 children followed in each school will experience at least one illness episode illness episodes warranting medical assessment over the 8-week observation window (this proportion likely under-estimates the prevalence of morbidity episodes we will observe during the data collection, which will take place during the rainy season, whereas the pilot occurred in a lower-morbidity season), and that they will seek medical care on average 70% of the time. Assuming an intra-cluster correlation of 0.35 (higher than the one estimated during the pilot), we are powered to detect a difference of 11.9 percentage point in the probability of seeking care between the SHP arm and the control arm. Appendix Table 1 shows that the range of minimal detectable effect size between 8.5-13.9 points, depending on the assumptions made about (1) the intra-cluster correlation, (2) the proportion of children seriously ill and (3) the proportion of healthcare utilisation in the control group. These MDEs are within the range of effects documented in randomised evaluations of interventions aiming to increase utilisation of care in LMICs. Performance-based financing in Rwanda, Zimbabwe, and Cambodia have found increases in facility-based service use of 7 to 9 percentage points (Basinga, Gertler et al. 2011, Van de Poel, Flores et al. 2016), while demand-side interventions combining financial incentives with improved access have generated effects of 27 percentage points (Thornton 2008). Our intervention primarily reduces the physical distance to care for children and families in rural areas, one of the most consistently documented barriers to healthcare utilisation in sub-Saharan Africa. As such, it is closest to studies looking at demand-side interventions reducing opportunity cost and distance such as Thornton (2008), which has been shown to generate among the largest demand responses in the literature. A 10-12 percentage point MDE therefore represents a conservative benchmark: it lies above the effects found in studies relying on information provision or modest financial nudges alone, yet comfortably within the range achieved by access-enhancing interventions operating in similarly supply-constrained environments. School attendance. We expect the school health programme to increase attendance in school (HA2). Earlier management of illnesses, school-based deworming, health education, and support to menstruating girls all help prevent common causes of absenteeism, such as untreated infections and poor menstrual hygiene, while the peer-buddy system ensures that children return to school more quickly after illness and that school staff are more attuned to learner absenteeism. For this outcome, the unit of analysis is the child. As described in section 3.4.2, our primary measure of attendance will be measured using the panel sample. At baseline, the intra-cluster correlation (ICC) of attendance at the school level using the panel-based method was 0.039 (controlling for stratification-cell fixed effects). In our power calculations, we test for the minimum detectable effect between the six comparisons we plan to make across treatment arms (e.g., Full Healthy Learners treatment vs. control, Full Healthy Learners treatment vs. deworming, and deworming vs. control). We calculate the minimum detectable effect size for each comparison assuming a two-sided significance level of α = 0.05, a conservative ICC of 0.05, and 80 percent power, accounting for clustering at the school level. We assume 89 Healthy Learners treatment schools (reflecting one treatment school that dropped out after baseline), 60 deworming schools, and 75 pure control schools, with an average of 58 sampled learners per school. We conduct these calculations across baseline attendance rates of 0.5, 0.6, and 0.7. Attendance rates measured in the panel survey are expected to be lower than those from government measures based on school registers due to survey attrition and the requirement that students be observed and present across multiple survey rounds. The study is powered to detect effects on attendance of approximately 4 to 6 percentage points. The results of this can be found in Appendix Table 2. Minimum detectable effects range from roughly 5.0 to 6.3 percentage points, with smaller MDEs at higher baseline attendance rates. Across comparisons, minimum detectable effects are largest for Deworming versus Control arms and smallest for comparisons involving the Full HL Treatment, reflecting differences in the number of schools across arms.