A compartmental model of the 1978 boarding-school influenza outbreak
The SIR model, with the infectious boys counted as the boys in bed, predicts 740 boys ill in an outbreak where 512 fell ill. A model with separate states for the infectious boys, the boys in bed and the convalescent boys predicts 524, without being told the total.
The influenza outbreak of January 1978 in an English boarding school is still used to present and to test epidemic models. Avilov et al.1 call it “famous” and report that earlier work “failed to fit classic SEIR epidemic models satisfactorily” to its data, SEIR models being SIR models with a stage of infected boys who are not yet infectious. The outbreak allows a check that most outbreak data do not. The school held 763 boys, the record2 gives the number in bed and the number convalescent on each day for two weeks, and it also gives how many fell ill in total, 512. A model fitted to the daily counts can therefore be checked against a total it never sees.
The SIR model with its infectious boys counted as the boys in bed
The SIR model divides a population into susceptible, infectious and recovered people. With the infectious boys taken to be the boys in bed, the height of the peak and the final number ill both follow from one number, the basic reproduction number, the average number of boys one sick boy infects while all the others can still catch the disease. The recorded peak of 298 boys in bed needs a value that would make 745 boys ill, and the recorded total of 512 needs one that would put only 70 in bed at the peak. No value gives both, and the fitted model predicts 740 boys ill.
Separate states for infectious, in bed and convalescent boys
A compartmental model counts the boys in each state of the illness and moves them from one state to the next. The S–E–I–B–C–R model follows each boy from susceptible through exposed, infectious, in bed and convalescent to recovered, so the two recorded series are states of the model itself. Of its versions, the one that best balances the fit to both series against its number of parameters predicts 524 boys ill, and between 491 and 558 for all parameter values the counts do not reject at the 95% level. It also finds that 24% to 37% of the boys were not susceptible at the start (95% confidence interval), through earlier immunity or little contact with the others.
The reproduction number and the effect of an intervention
The counts fix the length of a stay in bed, about three days, but not how contagious the outbreak was; the basic reproduction number lies anywhere from 3.9 to 9.3 (95% confidence interval). The difference matters for decisions. Halving transmission from 22 January would have spared 206 boys if the number was 4 and only 38 if it was 8, two values the counts cannot tell apart.
Forecasts from the first days
Fitted to the first eight days, the model is compatible with anything from 381 to 760 boys ill at the 95% level. The range narrows to 494 to 568 only with ten days of counts, four days after the peak.
Limitations
The data are one small closed population, with 14 daily readings taken from a published graph. The model assumes that every infected boy went to bed and that all boys mixed alike across the houses of the school. The model of Avilov et al.1, with infectious periods of fixed length written as delay differential equations, fits the daily counts more closely.
References
- Avilov, K. K. et al. The 1978 English boarding school influenza outbreak: where the classic SEIR model fails. Journal of the Royal Society Interface 21, 20240394 (2024). doi:10.1098/rsif.2024.0394
- Influenza in a boarding school. British Medical Journal 1, 586–590 (1978). PMC1603269