Reactive Case Detection and Treatment and Reactive Drug Administration for Reducing Malaria Transmission: A Systematic Review and Meta-Analysis

Abstract

Many countries pursuing malaria elimination implement "reactive" strategies targeting household members and neighbors of index cases to reduce transmission. These strategies include reactive case detection and treatment (RACDT; testing and treating those positive) and reactive drug administration (RDA; providing antimalarials without testing). We conducted systematic reviews of RACDT and RDA to assess their effect on reducing malaria transmission and gathered evidence about key contextual factors important to their implementation. Two reviewers screened titles/abstracts and full-text records using defined criteria (Patient = those in malaria-endemic/receptive areas; Intervention = RACDT or RDA; Comparison = standard of care; Outcome = malaria incidence/prevalence) and abstracted data for meta-analyses. The Grading of Recommendations, Assessment, Development, and Evaluations approach was used to rate certainty of evidence (CoE) for each outcome. Of 1,460 records screened, reviewers identified five RACDT studies (three cluster-randomized controlled trials [cRCTs] and two nonrandomized studies [NRS]) and seven RDA studies (six cRCTs and one NRS); three cRCTs comparing RDA to RACDT were included in both reviews. Compared with RDA, RACDT was associated with nonsignificantly higher parasite prevalence (odds ratio [OR] = 1.85; 95% CI: 0.96-3.57; one study) and malaria incidence (rate ratio [RR] = 1.30; 95% CI: 0.94-1.79; three studies), both very low CoE. Compared with control or RACDT, RDA was associated with non-significantly lower parasite incidence (RR = 0.73; 95% CI: 0.36-1.47; 2 studies, moderate CoE), prevalence (OR = 0.78; 95% CI: 0.52-1.17; 4 studies, low CoE), and malaria incidence (RR = 0.93; 95% CI: 0.82-1.05; six studies, moderate CoE). Evidence for reactive strategies' impact on malaria transmission is limited, especially for RACDT, but suggests RDA might be more effective.

Figure 1.

PRISMA diagram. RDA = reactive drug administration; RACDT = reactive case detection and treatment.

Figure 2.

Forest plot of comparison: reactive drug administration (RDA) versus no RDA on incidence of malaria infection. ¹Negative binomial regression with random effect (cluster level) and adjusted for first month of incidence, age, gender, household socioeconomic class, vector control, rainfall, enhanced vegetation index, and elevation.

Figure 3.

Forest plot of comparison: reactive case detection and treatment (RACDT) versus reactive drug administration (RDA) on prevalence of malaria infection. ¹The 95% CI upper limit presented here is artificially lower than in the published paper (odds ratio = 1.85, 95% CI: 0.96–20.00), because the authors of the Namibia trial calculated the effect size using marginal effects post-estimation (to account for reactive indoor residual spraying [IRS] in half the clusters) after a regression model, and Review Manager software can only accommodate balanced CIs. Effect size from (nonlinear) marginal effect post-estimation from generalized estimating equations (GEE) model using a logit function adjusted for reactive IRS, the interaction between reactive IRS and RDA, 2016 incidence of local cases, index case level and target population coverage for RDA or RACDT, response time, and co-interventions by Namibia Ministry of Health. Unadjusted effect size (from postestimation marginal effect of RDA from GEE model using a logit function adjusted for reactive IRS, the interaction between reactive IRS and RDA but no other covariates): 0.95 (95% CI: 0.48–33.3).

Figure 4.

Forest plot of comparison: reactive drug administration (RDA) versus no RDA/reactive case detection and treatment (RACDT) on prevalence of malaria infection. ¹Random effects logistic regression model adjusted for child age (in years), gender, household wealth from an asset index, rainfall, enhanced vegetation index, household elevation, and household protection by long-lasting insecticidal nets and indoor residual spraying (IRS). ²The 95% Cl lower limit is higher here than in the published paper (odds ratio = 0.54, 95% CI: 0.05–1.04) because the authors of the Namibia trial calculated the effect size using marginal effects post-estimation (to account for reactive IRS in half the clusters) after a regression model, and Review Manager software can only accommodate balanced CIs. Effect size from (nonlinear) marginal effect post-estimation from generalized estimating equations (GEE) model using a logit function with variables for RDA, reactive IRS, the interaction between reactive IRS and RDA, and adjusted for 2016 incidence of local cases. Unadjusted effect size (from post-estimation marginal effect of RDA from GEE model using a logit function with variables for RDA, reactive IRS, the interaction between reactive IRS and RDA but no other covariates): 1.05 (0.03–2.07). ³Random effects logistic regression (random effect for health facility) adjusted for age. Unadjusted odds ratio: 0.73 (95% CI: 0.27–1.94).

Figure 5.

Forest plot of comparison: reactive case detection and treatment (RACDT) versus reactive drug administration (RDA) on clinical malaria incidence. ¹Negative binomial analysis of monthly facility cases (random intercept for facility); adjusted for previous month’s cases, normalized difference vegetation index, precipitation, altitude, nighttime light, number of RDTs done each month, and seasonality (Fourier term). Unadjusted estimate: 0.93 (0.77–2.00). ²The 95% Cl upper limit is lower here than in the published paper (1.41, 95% CI: 0.83–4.55) because the authors of the Namibia trial calculated the effect size using marginal effects post-estimation (to account for reactive indoor residual spraying [IRS] in half the clusters) after a regression model, and Review Manager software can only accommodate balanced CIs. Estimate from (nonlinear) marginal effect post-estimation from a negative binomial model with offset for cluster-level person time; adjusted for reactive vector control, interaction between RDA and reactive vector control, 2016 incidence of local cases, index case level and target population coverage for RACDT or RDA, response time, and cointerventions by Namibia Ministry of Health. Unadjusted marginal effects from post-estimation (from unadjusted negative binomial model with terms for RACDT, reactive IRS, and the interaction between the two, with offset for cluster-level person time): 1.22 (0.73–3.85). ³Negative binomial regression model of local cases with offset for person-time and adjusted for baseline (2014–2015) incidence of local cases. Unadjusted estimate: 0.94 (95% CI: 0.51–1.75).

Figure 6.

Forest plot of comparison: reactive drug administration (RDA) versus no RDA/reactive case detection and treatment (RACDT) on clinical malaria incidence. ¹Negative binomial analysis of monthly facility cases (random intercept for facility); adjusted for previous month’s cases, normalized difference vegetation index (NDVI), precipitation, altitude, nighttime light, number of rapid diagnostic tests done each month, and seasonality (Fourier term). Unadjusted estimate: 1.08 (95% CI: 0.78–1.49). ²Negative binomial difference-in-differences model (one pre- and one post-time point), adjusted for prior month’s cases, calendar month, rainfall, and EVI monthly anomalies. ³The 95% CI lower limit is higher here than in the published paper (rate ratio = 0.71 (95% CI: 0.22–1.20). Effect size from (nonlinear) marginal effect post-estimation from a negative binomial model with offset for cluster-level person time; variables for RDA, reactive vector control, interaction between RDA and reactive vector control, and adjusted for 2016 incidence of local cases. Unadjusted marginal effects from post-estimation (from unadjusted negative binomial model with terms for RACDT, reactive indoor residual spraying, and the interaction between the two, with offset for cluster-level person time): 0.82 (0.26–1.37). ⁴Poisson regression model adjusted for age. Unadjusted estimate from a logistic regression model (with a random effect for cluster): 1.04 (95% CI: 0.57–1.91). ⁵Negative binomial regression model of local cases with offset for person-time and adjusted for baseline (2014–2015) incidence of local cases. Unadjusted estimate: 1.06 (95% CI: 0.57–1.98).

Figure 7.

Summary of reactive case detection and treatment (RACDT) and reactive drug administration (RDA) outcomes and Grading of Recommendations, Assessment, Development, and Evaluations for each. RCT = randomized controlled trial; NRS = nonrandomized study. ¹Rated not serious on risk of bias, inconsistency, indirectness; rated serious on imprecision. ²Rated not serious on risk of bias, inconsistency; rated serious on indirectness, imprecision. ³Rated not serious on risk of bias, inconsistency, imprecision; rated serious on indirectness. ⁴Rated serious on risk of bias; rated not serious on inconsistency, indirectness, imprecision. ⁵Rated not serious on risk of bias, inconsistency; rated very serious on indirectness; rated serious on imprecision. ⁶Rated not serious on risk of bias, inconsistency; rated very serious on indirectness; rated serious on imprecision.