Situation: A public health nurse in a rice-farming municipality reviews a published prospective cohort study of farmers before planning leptospirosis prevention with the local farmers' association. The association wants to know whether giving free rubber boots to farmers lowers the incidence of leptospirosis. Which design would give the STRONGEST evidence?

To test whether an intervention works, the strongest single-study design is an experiment in which the investigator assigns the intervention at random. Randomizing whole barangays (a community or cluster trial) balances known and unknown confounders between the groups. Observational designs cannot rule out that farmers who choose boots differ in other ways.

Study Design for Evaluating an Intervention

The question asks which design provides the strongest evidence for whether free rubber boots reduce leptospirosis incidence. The correct answer is a cluster-randomized trial assigning barangays to receive boots.

When the goal is to test whether an intervention causes a change in outcome, randomization is the key feature that separates strong evidence from weak evidence. Random assignment—whether of individuals or of entire communities such as barangays—distributes both known and unknown confounding variables evenly between the intervention and control groups. This means any observed difference in leptospirosis incidence can be more confidently attributed to the boots rather than to other differences between farmers.

Confounding is the central problem in the other three designs. Farmers who choose to wear boots, or who already own boots, may differ systematically from those who do not. They may have higher income, better health literacy, greater concern about infection, or different work practices. These differences—not the boots themselves—could explain lower disease rates. Observational designs cannot fully separate the effect of the boots from the effect of the person who chooses to wear them.

Design	Investigator assigns intervention?	Controls confounding?	Evidence strength for causation
Cluster-randomized trial (barangays randomized)	Yes, by random assignment	Strong—balances known and unknown confounders	Strongest
Survey of current boot use and past leptospirosis	No—exposure is self-selected	Weak—cross-sectional, cannot establish temporality	Weakest
Case-control study	No—exposure is recalled after outcome	Moderate—matching possible but recall bias and unmeasured confounders remain	Moderate
Prospective cohort of farmers who choose to wear boots	No—exposure is self-selected	Moderate—can measure some confounders but not unknown ones	Moderate

Key point! Randomization is what makes a design experimental rather than observational. A cluster trial randomizing barangays is the strongest feasible design here because individual randomization of farmers within the same community would be impractical—boot use could spread between neighbors, contaminating the control group.

The leptospirosis literature supports the relevance of environmental exposure. Leptospira are shed in the urine of reservoir hosts such as rats, and humans become infected through direct or indirect contact with contaminated water or soil ^[1]. In rice-farming settings, farmers work in flooded fields where prolonged skin contact with contaminated water is a major exposure route. Rubber boots create a physical barrier that plausibly interrupts this transmission pathway. However, plausibility alone is not proof of effectiveness—only a randomized comparison can demonstrate that the boots actually reduce incidence.

A Cochrane review on antibiotic prophylaxis for leptospirosis illustrates why observational evidence is insufficient for prevention decisions. The review authors noted that clinical benefits of prophylactic antibiotics remain uncertain despite biological plausibility, because high-quality randomized evidence is limited ^[2]. The same logic applies to boots: without randomization, self-selected boot users may simply be at lower risk for reasons unrelated to the boots themselves.

Watch out! A prospective cohort study (option 4) is stronger than a survey or case-control study for establishing temporality, but it still cannot rule out confounding by indication—the tendency for people who adopt protective behaviors to differ in other health-promoting ways. Only randomization breaks this link between the choice to use an intervention and other personal characteristics.

In a cluster-randomized trial, the unit of randomization is the barangay, not the individual farmer. This design is appropriate when the intervention is delivered at the community level and when individual randomization risks contamination. The trade-off is that cluster trials require more participants to achieve the same statistical power, because outcomes within a cluster tend to be correlated. Despite this practical challenge, the design remains the strongest single-study approach for answering the farmers' association's question about whether free boots lower leptospirosis incidence.