baseIt_RM {flubase}R Documentation

Baseline free of influenza epidemic effects: iterative procedure with cyclical regression model

Description

This function estimates the mortality (or other indicator) baseline, free of influenza epidemics, using an iterative procedure. The baseline for each flu-year (week 27 to week 26 or month 7 to month 6) is estimated from fitting a cyclical regression model to a training set of previous flu-years, without the periods with excess deaths associated with influenza epidemics. In each iteration the model identifies the periods with excess deaths as those weeks (months) above the 95 CI of the baseline.

Usage

baseIt_RM(nod, todeath, epi, flu_year, ny, tb, te, pe, ni = 5, train = 5, per)

Arguments

nod a vector with the number of deaths (or other indicator) by week or month
todeath a vector that contains the time index (week or month number)
epi a vector that indicates if the week or month belongs to the epidemic period, in which case epi=1. Otherwise, epi=0
flu_year a vector that indicates the flu year. It is an index for the set of 52 weeks or 12 months, that initiate at week 27 and ends at week 26 of the next civil year, or iniate at month 7 and ends at month 6 of the next civil year, depending on the time unit of data.
ny number of years in study
tb initial week (tb=48) or month (tb=12) of the fixed epidemic period
te final week (te=17) or month (te=4) of the fixed epidemic period
pe pe = 0 if the user provides the epidemic periods in the epi parameter; otherwise if pe = 1 the function uses a fixed period from week 47 to week 17 or from month 12 to month 4.
ni represents the number of inicial flu-years to inicialize the the iterative procedure, by default is equal to 4
train is the set of flu years used to predict the baseline for the flu year i+1, by default is equal to 5
per per=52 if the data is weekly or per=12 if the data is monthly

Details

The objective of this function is to estimate a mortality baseline without the effect of influenza epidemics. With this purpose the function starts fitting a cyclical regression model to the first ni flu years of mortality time series after excluding the epidemic periods, i.e. the values of nod when epi=0. Then it engages in an iterative procedure where in each iteration the function forecasts the baseline of mortality without the effect of influenza epidemics for flu year i+1 using the model fitted to the flu years i-train+1 to i. In each iteration the function also identifies the periods with excess mortality associated to influenza epidemics in year i+1, as those where the observed mortality initiate with two consecutive observations above the 95 CI of the forecasted baseline and ends with two observations bellow the 95 CI of the forecasted baseline. These periods are returned in da variable.

Value

The function will return a list

baseIt_RM$beta0 containing the mortality (or other indicator) baseline without the effect of influenza epidemics
baseIt_RM$beta_up containg the upper 95 CI of the baseline
baseIt_RM$da a dummy variable indicating the periods with excess deaths associated with the ocurred influenza epidemics

Note

The ni must be minor or equal to train.

Author(s)

Nunes B, Natario I and Carvalho L.

References

Nunes B, Natario I, Carvalho L. Time series methods for obtaining excess mortality. Submitted to Statistical Methods in Medical Research (2009).

Serfling RE Methods for Current Statistical Analysis of Excess Mortality Pneumonia-Influenza Deaths Public Heath Reports 1963; 78 6:494 506.

Lui K-J and Kendal A.P. Impact of influenza epidemics on mortality in the united states from october 1972 to may 1985. American Journal of Public Health 1987; 77(6):712 716.

L. Simonsen, M.J. Clarke, D- Williamson, D.F. Stroup, N.H. Arden and L.B. Schonberger The impact of influenza vaccination on seasonal mortality in the US elderly population. American Journal of Public Health 1997; 87(12):1994 1950.

See Also

baseIt_SA, baseSA, baseIt_RM, flubase


[Package flubase version 1.0 Index]