To the Editor: Since 2004, when my colleagues and I publisheda report estimating the number of excess deaths resulting fromthe invasion of Iraq,1 we have made further assessments, publishedin 2006.2 The report on the Iraq Family Health Survey (IFHS)(Jan. 31 issue)3 contains some findings that are similar tothose of our 2006 study but some that differ. The IFHS reportestimated a very low crude mortality rate before the invasionas compared with the rate we calculated (3.17 vs. 5.5). Arguably,the increases in mortality reported in both studies for theperiod after the invasion were more similar: 6.01÷3.17=1.9for the IFHS versus 13.2÷5.5=2.4 for our study, a differenceof 21%.3 It is unfortunate that the IFHS focused only on violence-relatedmortality.
IFHS modeling estimated that only 38% of deaths were unreported,but comparisons with rates of death in Iran and Syria suggestunderreporting of 55% and 70%, respectively. The growth balancemethod used in the IFHS is untested in volatile populations.How did the omission of data from approximately 10% of the mostviolent parts of Iraq produce a relatively small increase inthe width of the confidence interval? Most sources (the IraqBody Count,4 the Baghdad morgue, and the Pentagon) show moreviolence in 2006 than in 2003 or 2004, which is not evidentin the IFHS report. However, these data suggest that our 2004estimate of approximately 100,000 excess deaths was too low.
Gilbert M. Burnham, M.D., Ph.D. Johns Hopkins Bloomberg School of Public Health Baltimore, MD 21205 gburnham{at}jhsph.edu
References
Roberts L, Lafta R, Garfield R, Khudhairi J, Burnham G. Mortality before and after the 2003 invasion of Iraq: cluster sample survey. Lancet 2004;364:1857-1864. [CrossRef][Web of Science][Medline]
Burnham G, Lafta R, Doocy S, Roberts L. Mortality after the 2003 invasion of Iraq: a cross-sectional cluster sample survey. Lancet 2006;368:1421-1428. [CrossRef][Web of Science][Medline]
Iraq Family Health Survey Study Group. Violence-related mortality in Iraq from 2002 to 2006. N Engl J Med 2008;358:484-493. [Free Full Text]
To the Editor: As the Perspective article accompanying the IFHSreport points out,1 estimating the number of people killed asa result of a conflict is challenging. The media has widelyreported the large difference between the rates of violence-relateddeaths reported in the IFHS and the rates of death from allcauses reported by Burnham et al. in 2006.2 The IFHS study groupreported the primary outcome of violence-related mortality andfound an estimated 150,000 deaths between 2003 and 2006, whereasBurnham et al. appropriately reported the rate of death fromall causes during a similar period and found an estimated 654,000deaths. The majority of deaths that result from any conflictare attributable not to violence but to the complex dilapidationof the normally protective public health infrastructure.3 Focusingon violence alone ignores many of the deaths that have occurredduring this invasion. Using the crude mortality rates in theIFHS report, the actual excess mortality in Iraq between 2003and 2006 was approximately 433,000 (95% confidence interval[CI], 354,000 to 523,000). Indeed, absent from the IFHS reportis an acknowledgment that the combined totals actually approachthose of the 2006 study by Burnham et al.
Edward J. Mills, Ph.D., L.L.M. British Columbia Centre for Excellence in HIV/AIDS Vancouver, BC V6R 2H4, Canada emills{at}cihhrs.org
Frederick M. Burkle, M.D., M.P.H. Harvard Humanitarian Initiative Cambridge, MA 02138
References
Brownstein CA, Brownstein JS. Estimating excess mortality in post-invasion Iraq. N Engl J Med 2008;358:445-447. [Free Full Text]
Burnham G, Lafta R, Doocy S, Roberts L. Mortality after the 2003 invasion of Iraq: a cross-sectional cluster sample survey. Lancet 2006;368:1421-1428. [CrossRef][Web of Science][Medline]
Spiegel PB, Le P, Ververs MT, Salama P. Occurrence and overlap of natural disasters, complex emergencies and epidemics during the past decade (1995-2004). Confl Health 2007;1:2-2. [CrossRef][Medline]
To the Editor: The estimated mortality rate in Iraq before theinvasion, as reported by the IFHS study group, was lower thanthe estimated rates in neighboring countries1 and lower thanhalf the rates estimated from other sources.1,2,3 Although thesurvey's large size allows for more precision, systematic error(bias) is not avoided. The authors acknowledge the difficultiesinvolved in surveying high-mortality governorates (HMGs), butreliable figures for these areas are essential for obtainingan overall estimate of violence-related mortality.
Discrepancies with previous estimates of the violence-relatedmortality rate in Iraq — a difference by a factor of almost10 for HMGs (Table 1) — could be due to the low proportionof violent deaths attributed to HMGs as compared with the populationin these areas (31% vs. 38%); the fact that the IFHS imputedmissing data for HMGs using the Iraq Body Count, which probablyunderweights the HMGs as a result of publication bias; and useof a very long questionnaire, which has been shown to lead tounderestimates of mortality.4
Table 1. Geographic Distribution of Violence-Related Deaths in Iraq during the Post-Invasion Period (May 2003–June 2006), According to the IFHS and Burnham et al.
A reliable estimate of the violence-related mortality rate inHMGs is urgently needed to clarify the conflict-related deathtoll in Iraq.
Francisco J. Luquero, M.D., M.P.H. Rebecca F. Grais, Ph.D. Epicentre 75011 Paris, France
Dr. Luquero reports receiving grant support from the EuropeanCentre for Disease Prevention and Control.
Burnham G, Lafta R, Doocy S, Roberts L. Mortality after the 2003 invasion of Iraq: a cross-sectional cluster sample survey. Lancet 2006;368:1421-1428. [CrossRef][Web of Science][Medline]
Roberts L, Lafta R, Garfield R, Khudhairi J, Burnham G. Mortality before and after the 2003 invasion of Iraq: cluster sample survey. Lancet 2004;364:1857-1864. [CrossRef][Web of Science][Medline]
To the Editor: The IFHS study group made a serious error byassuming no spatial bias in the Iraq Body Count database. TheIraq Body Count project does not purport to be a random sample;it reflects the distribution of reporters as much as it doesthe distribution of violence. The IFHS spatial results alignclosely with those of the Iraq Body Count not because of thesurvey findings (the survey group did not visit the most dangerousclusters) but because of the assumptions they imported fromthe spatial pattern in the Iraq Body Count. If we assume thatthe spatial distribution in the study reported by Burnham etal.1 is correct (it remains the only study to scientificallyinvestigate this distribution), that factor alone could doublethe number of violence-related deaths found by IFHS to 300,000.The authors acknowledge and attempt to correct for underreportingof deaths from nonviolent causes, but they make no allowancefor the more serious underreporting of violence-related deathsto government-affiliated survey takers. This leads the IFHSto implausibly conclude that less than one third of excess deathswere due to violence. When these two sources of error are takentogether, the IFHS results are easily in line with the findingof more than 600,000 violent deaths in the study by Burnhamet al.
Timothy R. Gulden, Ph.D. University of Maryland School of Public Policy College Park, MD 20742 tgulden{at}umd.edu
References
Burnham G, Lafta R, Doocy S, Roberts L. Mortality after the 2003 invasion in Iraq: a cross-sectional cluster sample survey. Lancet 2006;368:1421-1428. [CrossRef][Web of Science][Medline]
The authors reply: Our estimate of violent deaths in Iraq fromMarch 2003 to June 2006 was based on the 2006–2007 IFHS,a cross-sectional, nationally representative survey of 9345households. We reported only the analysis of violence-relatedmortality, not our analysis of deaths due to other causes.
Substantial underreporting of deaths is common in householdsurveys because of recall bias, the effects of migration, andmissing households. The security risks in Iraq add to the reportingproblems. The geographic heterogeneity of violence-related deathrates may have further affected reported deaths, even though971 clusters were sampled.
The preinvasion crude mortality rate of 3.17 (95% CI, 2.70 to3.75) in the IFHS report is lower than the rate of 5.5 (95%CI, 4.3 to 7.1) reported by Burnham et al. Because the levelof underreporting is almost certainly higher for deaths in earliertime periods, we did not attempt to estimate excess deaths.The excess deaths reported by Burnham et al. included only 8.2%of deaths from nonviolent causes, so inclusion of these deathswill not increase the agreement between the estimates from theIFHS and Burnham et al.
We imputed data for missing clusters in the Anbar province amongthe HMGs, and we included uncertainty in this imputation; itcannot account for the 10-fold difference between our ratesand those reported in other HMGs. Burnham et al. selected onlya few clusters in each of the HMGs, which may not be representativeof those governorates. We estimated that 45% of violence-relateddeaths were in HMGs after the adjustment, as compared with 36%of the Iraqi population. This may even be an overestimate ifthe reports collected by the Iraq Body Count project concentrateon high-impact events closer to main cities.
Although the estimated number of violence-related deaths inthe IFHS is approximately three times higher than those reportedby the Iraq Body Count project, the results are consistent withthe Iraq Body Count trends and distribution, based on collationsof press reports for civilian casualties. As indicated in Table4 of our report, the IFHS and the Iraq Body Count both showeda drop and a subsequent increase in violence-related deaths,but not a doubling and then more than a fourfold increase forthe same time periods as reported by Burnham et al.
To reach the 2005–2006 death rate of more than 900 perday, estimated by Burnham et al., the IFHS would have had tomiss nearly 90% of violence-related deaths. It is unlikely thata small survey with only 47 clusters has provided a more accurateestimate of violence-related mortality than a much larger surveysampling of 971 clusters. We may never know with any accuracythe effect of the conflict in Iraq on mortality, but all theevidence points to a high level of deaths due to violence.
Mohamed M. Ali, Ph.D. Colin Mathers, Ph.D. J. Ties Boerma,Ph.D. World Health Organization CH-1211 Geneva 27, Switzerland alim{at}emro.who.int
Previous
