Risk factors and patterns of household clusters of respiratory viruses in rural Nepal

Scott, E. M.; Magaret, A.; Kuypers, J.; Tielsch, J. M.; Katz, J.; Khatry, S. K.; Stewart, L.; Shrestha, L.; LeClerq, S. C.; Englund, J. A.; Chu, H. Y.

For each examination of viral transmission, initial viral infections were those preceded by a 14-day period without infections of that virus type in a household. Index or transmitting cases were established by identifying the individual(s) with the earliest reported respiratory symptoms prior to a virus-positive swab. We computed the incidence of secondary household illness cases following each initial viral infection. We defined transmission events as observed infections of the same virus in another member of the same household within the subsequent 14 days. The 14-day risk period began on the first day of criteria respiratory symptoms associated with virus-positive specimen collection in the week preceding the virus-positive swab in the index case. Each initial infection and its corresponding risk period comprise a single data point in the regressions and include subsequent infections and time at risk from all household members reporting symptoms over that time period. To assess the risk factors for the incidence of secondary cases of these initial illnesses within households, we used generalised estimating equations, accounting for the potential similarity among repeated initial infections within households. The outcome was the number of secondary cases, with a fixed offset of the log time at risk; a log link was used to estimate the incidence rate ratios (IRRs). Multivariable regression was performed by first including all measures significant in the univariable analysis at P < 0.1 and then performing backward elimination to select a final model. Measures were retained in final multivariable regression if significant at P < 0.05 or if had a substantial impact (⩾10% shift in estimate) on other significant covariates. A single index case type was included in the multivariable model as the index case was coded such that one index case type was compared to all other index cases.