The limitations of our study include potential selection bias and the lack of a population-based estimate due to the sampling from IDP camps. Although we identified all pre-tsunami family members by interviewing the householders, we examined only the households that had been accommodated in the IDP camps. Thus, the mortality presented in this study is probably higher than the population-based statistics that include people who did not take refuge in the IDP camps but stayed in their own houses after the tsunami, which obviously would have suffered less destruction.
It is also possible that our sample population is an even more vulnerable group among all of the IDP affected by the tsunami. This is because by the time the survey was conducted, relatively affluent families might have moved out from IDP camps and resettled in other locations such as a relative's house. If it is the case, the variation of the exposure measurements were likely to be reduced in our study sample, potentially leading to the decreased effects of the risk factors on mortality. Nevertheless, considering the fact that our sample preserved its population structure representing the national statistics, we believe that our study population is still useful for identifying the risk factors of mortality especially for demographic factors.
We observed unexpected findings for some socio-economic variables. Decreased mortality was observed among families with the lowest educational level and apparently there was no association between household income levels and mortality. It has been generally recognized that various socio-economical indicators are the important determinants in predicting who are the most vulnerable during disasters. A question may be raised as to whether we can conclude that our results are inconsistent with this general perception; possibly reflecting the non-discriminative nature of the tsunami disaster. However, we cannot draw a conclusive statement from our study because of the limitations set-out below.
There might have been some residual confounding effects behind some of the associations which were not totally controlled. For example, the families in the lowest educational category tended to have a higher age distribution of the family members, which probably reflects the long-term improvement of school enrolment rate over the past decades in Sri Lanka. Because high mortality was observed among young children aged less than 10 years, families whose members were older in age and often without small children, might have experienced decreased mortality.
Although we have conducted a multivariate analysis, it is possible that the confounding effect by age was not totally controlled because of the difference in the measurement level (educational level was measured at household level while age at individual level).
Another important limitation could be the lack of enough variation in exposures due to the sampling from possibly the most vulnerable IDP as noted above. Although the income level ranged from zero to more than 15,000, the majority falls into the categories of 1–2999 and 3000–5999 Sri Lankan Rupees and few households were in the highest category. This distribution of income level seems low comparing previous available data. According to the Household Income and Expenditure Survey 2002/3 by Department of Census and Statistics, the median and mean household income for the Eastern Province (where our study area is located) were 5500 and 7640 respectively and around 20% of the sample households reported more than 10,000 Sri Lankan Rupees of income.
Therefore, it is possible that our sample population was a somewhat homogeneous population from relatively low socio-economic groups. This meant that the effect of socio-economic indicators on mortality did not appear to be substantial, i.e. there were reduced effects due to the lack of variation in exposures. Considering these points, our study could not provide a decisive conclusion on the correlation between some socio-economic indicators and mortality.
This study revealed a significantly high mortality rate in women, children and the elderly. A similar gender-age mortality pattern has been reported in other disaster settings. In both the 1999 Taiwan earthquake [7, 8] and the 1988 Armenian earthquake [9], the elder population was particularly at risk of death [10], and there was a high mortality rate among women and young children. In the 1995 Great Hanshin-Awaji Earthquake in Japan, a significantly high mortality rate was reported among the elderly, especially those with physical disabilities [11].
By contrast, previous studies on floods have revealed clearly different mortality patterns. In an extensive study on flood-related deaths in Europe and the United States, middle-aged men were found to be the most vulnerable population [12]. A study in Australia also reported that 80% of flood deaths occurred among the male population [13]. Both studies suggested that risky behaviour, such as trying to swim across rivers or using motor vehicles to flee, resulted in the increased mortality of this population.
The findings of our study indicate that the age-gender mortality pattern of the tsunami victims is similar to that of earthquakes but very different to that of floods. This clear difference in the high-risk groups observed in different types of disasters highlights the need for epidemiological studies to precisely identify the vulnerable groups unique to each disaster setting, and to provide information for subsequent disaster management and planning. We believe that our study contributes to the epidemiological evidence on tsunami-related mortality, which has not been well described to date.
In planning reconstruction aid activities, the impact of the change in population structure on the affected community should be thoroughly considered. The loss of the middle-aged female population may have a particularly significant impact, because these individuals are the primary caretakers of their children and families, as highlighted by the officials from WHO and UNICEF [14]. The nutritional status and morbidity among the surviving children, including psychological illness, should be carefully monitored. The incidence of certain conditions among the surviving adult population, especially cardiovascular disease, as already noted in our study population [5], or mental disorders including suicidal tendencies, demands further attention because the long-term risks of these conditions have been pointed out in other disaster settings [15–17]. Although not imminent, social instability and the danger of an increase in sexually transmitted diseases, including HIV/AIDS, also deserve attention because the male-female ratio increased from 97.8:100 to 108.3:100 (before and after the tsunami) among people under 50 years of age in our study sample.
The consequences of disasters are very complex. Disaster epidemiology can play an important role in describing the adverse health effects and identifying the vulnerable groups during and after disasters. It enables relief and reconstruction activities to be more focused, relevant and efficient. Continuing research is needed to monitor and evaluate the long-term impact of the tsunami disaster on human health.
