Intended for healthcare professionals


Excess winter mortality: influenza or cold stress? Observational study

BMJ 2002; 324 doi: (Published 12 January 2002) Cite this as: BMJ 2002;324:89
  1. G C Donaldson, senior research associate,
  2. W R Keatinge, emeritus professor (w.r.keatinge{at}
  1. Medical Sciences, Division of Biomedical Sciences, Queen Mary and Westfield College, London E1 4NS
  1. Correspondence to: W R Keatinge
  • Accepted 10 July 2001

Epidemics of influenza are associated with increases in mortality and morbidity.1 Health professionals and the media, therefore, have often focused their attention on influenza as a cause of increased mortality and demands on health services in winter. Cold weather alone causes striking short term increases in mortality, mainly from thrombotic and respiratory disease.2 Non-thermal seasonal factors such as diet may also affect mortality.3 The increases in mortality are greater in London than in regions surveyed in continental Europe.4 We used multiple regression to assess the proportion of excess winter mortality that was attributable to influenza in south east England.

Methods and results

A daily record was kept of deaths that occurred in south east England from 1970 to 1999 for all causes and for influenza. We obtained daily estimates of population by linear regression from mid-year values (17.2×106 in 1971 and 18.4×106 in 1998) and used them to calculate mortalities. We used the maximum and minimum temperature at Heathrow Airport each day to obtain the mean. Temperature was lagged three days to give the steepest relation between temperature and mortality.2 Total mortality each year related to cold was obtained as the sum of excess daily mortalities (per million). Excess daily mortalities were mortalities that occurred below the temperature—in a 3°C band—at which mortality was lowest, compared with mortality in that band. The mean temperature of the lower limit of the band over the 30 years was 19.0°C (95% confidence interval 18.2°C to 19.8°C).

Influenza epidemics cause deaths additional to those registered as being due to influenza, such as deaths caused by arterial thrombosis. Therefore, we estimated total mortality related to influenza. Daily mortality was the dependent variable; we used mean registered deaths due to influenza over the period five days before and after the index day as the explanatory variable, and temperature at three day lag as confounding variable, with a linear time trend term. The regression used daily data in the linear portion of the temperature-mortality relation (range 0-15°C), pooled for 1970-99. To eliminate autoregression2 without distorting quantitative relations, the regression used a train of data spaced at 15 day intervals, starting 1 January 1990. The regression was repeated for similar trains starting on each consecutive day from 2 to 15 January to give 15 estimates of the mean of total influenza related mortality per recorded death from influenza. The 15 values averaged 5.1 (95% confidence interval 4.4 to 5.9) per million. We used this figure to calculate annual mortality related to influenza. Multicolinearity was acceptably low (variance inflation factor 1.02).

The annual rate of deaths caused by influenza has declined with time (figure). Mortality increased sharply during some epidemics, but even during the worst epidemic, in 1976, only 143 deaths per million were registered as due to influenza. Total influenza related deaths that year were calculated as 729 per million, less than half the total of excess winter deaths (2308 per million). Over the past 10 years, deaths registered as due to influenza averaged 5.01 per million per year, and annual influenza related deaths averaged 29.9 per million, or 2.4% (2.0% to 2.7%) of 1265 annual excess winter deaths per million.


Mortality due to influenza and total excess winter mortality in south east England, 1970-99. All results are per million and the vertical bars represent 95% confidence intervals


Of 1265 annual excess winter deaths per million over the past 10 years, 2.4% were due to influenza either directly or indirectly. The decline in influenza related deaths is probably due to immunisation and to a reduction in the number of new viral strains. With influenza causing such a small proportion of excess winter deaths, measures to reduce cold stress offer the greatest opportunities to reduce current levels of winter mortality. Warm housing is important but it can coexist with high winter mortality,5 and outdoor cold stress has been independently associated with high excess winter mortality.4 Campaigns to reduce exposure to cold outdoors provide obvious scope for future preventive action.


The Office for National Statistics supplied mortality and population data and the Royal Meteorological Office supplied the temperature data.

Contributors: Both authors designed the study, assessed the data, and wrote the paper. GD computed the data and WRK drafted the paper. Both are guarantors for the paper.


  • Funding EU Biomed grant.

  • Competing interests None declared.


  1. 1.
  2. 2.
  3. 3.
  4. 4.
  5. 5.
View Abstract