Monitoring health interventions--who's afraid of LQAS?

Lot quality assurance sampling (LQAS) is used to evaluate health services. Subunits of a population (lots) are accepted or rejected according to the number of failures in a random sample (N) of a given lot. If failures are greater than decision value (d), we reject the lot and recommend corrective actions in the lot (i.e. intervention area); if they are equal to or less than d, we accept it. We used LQAS to monitor coverage during the last 3 days of a meningitis vaccination campaign in Niger. We selected one health area (lot) per day reporting the lowest administrative coverage in the previous 2 days. In the sampling plan we considered: N to be small enough to allow us to evaluate one lot per day, deciding to sample 16 individuals from the selected villages of each health area, using probability proportionate to population size; thresholds and d to vary according to administrative coverage reported; α ≤5% (meaning that, if we would have conducted the survey 100 times, we would have accepted the lot up to five times when real coverage was at an unacceptable level) and ß ≤20% (meaning that we would have rejected the lot up to 20 times, when real coverage was equal or above the satisfactory level). We classified all three lots as with the acceptable coverage. LQAS appeared to be a rapid, simple, and statistically sound method for in-process coverage assessment. We encourage colleagues in the field to consider using LQAS in complement with other monitoring techniques such as house-to-house monitoring.

Multidrug resistance among new tuberculosis cases: detecting local variation through lot quality-assurance sampling.

BACKGROUND: Current methodology for multidrug-resistant tuberculosis (MDR TB) surveys endorsed by the World Health Organization provides estimates of MDR TB prevalence among new cases at the national level. On the aggregate, local variation in the burden of MDR TB may be masked. This paper investigates the utility of applying lot quality-assurance sampling to identify geographic heterogeneity in the proportion of new cases with multidrug resistance. METHODS: We simulated the performance of lot quality-assurance sampling by applying these classification-based approaches to data collected in the most recent TB drug-resistance surveys in Ukraine, Vietnam, and Tanzania. We explored 3 classification systems- two-way static, three-way static, and three-way truncated sequential sampling-at 2 sets of thresholds: low MDR TB = 2%, high MDR TB = 10%, and low MDR TB = 5%, high MDR TB = 20%. RESULTS: The lot quality-assurance sampling systems identified local variability in the prevalence of multidrug resistance in both high-resistance (Ukraine) and low-resistance settings (Vietnam). In Tanzania, prevalence was uniformly low, and the lot quality-assurance sampling approach did not reveal variability. The three-way classification systems provide additional information, but sample sizes may not be obtainable in some settings. New rapid drug-sensitivity testing methods may allow truncated sequential sampling designs and early stopping within static designs, producing even greater efficiency gains. CONCLUSIONS: Lot quality-assurance sampling study designs may offer an efficient approach for collecting critical information on local variability in the burden of multidrug-resistant TB. Before this methodology is adopted, programs must determine appropriate classification thresholds, the most useful classification system, and appropriate weighting if unbiased national estimates are also desired.

Clustered lot quality assurance sampling to assess immunisation coverage: increasing rapidity and maintaining precision.

OBJECTIVE: Vaccination programmes targeting disease elimination aim to achieve very high coverage levels (e.g. 95%). We calculated the precision of different clustered lot quality assurance sampling (LQAS) designs in computer-simulated surveys to provide local health officers in the field with preset LQAS plans to simply and rapidly assess programmes with high coverage targets. METHODS: We calculated sample size (N), decision value (d) and misclassification errors (alpha and beta) of several LQAS plans by running 10 000 simulations. We kept the upper coverage threshold (UT) at 90% or 95% and decreased the lower threshold (LT) progressively by 5%. We measured the proportion of simulations with < or =d individuals unvaccinated or lower if the coverage was set at the UT (pUT) to calculate beta (1-pUT) and the proportion of simulations with >d unvaccinated individuals if the coverage was LT% (pLT) to calculate alpha (1-pLT). We divided N in clusters (between 5 and 10) and recalculated the errors hypothesising that the coverage would vary in the clusters according to a binomial distribution with preset standard deviations of 0.05 and 0.1 from the mean lot coverage. We selected the plans fulfilling these criteria: alpha < or = 5% beta < or = 20% in the unclustered design; alpha < or = 10% beta < or = 25% when the lots were divided in five clusters. RESULT: When the interval between UT and LT was larger than 10% (e.g. 15%), we were able to select precise LQAS plans dividing the lot in five clusters with N = 50 (5 x 10) and d = 4 to evaluate programmes with 95% coverage target and d = 7 to evaluate programmes with 90% target. CONCLUSION: These plans will considerably increase the feasibility and the rapidity of conducting the LQAS in the field.