Tropical Medicine and International Health doi:10.1111/j.1365-3156.2009.02301.x volume 14 no 9 pp 969–976 september 2009

Epidemiological studies of the ‘non-specific effects’ of : I – data collection in observational studies

Paul E. M. Fine1, Thomas N. Williams2,3,4, Peter Aaby5,6, Karin Ka¨llander7,8, Lawrence H. Moulton9, Katie L. Flanagan10, Peter G. Smith1 and Christine S. Benn5,6 on behalf of the Working Group on Non-specific Effects of Vaccines*

1 London School of Hygiene & Tropical Medicine, London, UK 2 Kenya Medical Research Institute Centre for Geographic Medicine Research Coast, Kilifi District Hospital, Kilifi, Kenya 3 Nuffield Department of Medicine, John Radcliffe Hospital, Oxford, UK 4 Department of Paediatrics, John Radcliffe Hospital, Oxford, UK 5 Proje´cto de Sau´ de Bandim, Indepth Network, Bissau, Guinea-Bissau 6 Bandim Health Project, Danish Epidemiology Science Centre, Statens Seruminstitut, Copenhagen, Denmark 7 Division of International Health, Karolinska Institutet, Stockholm, Sweden 8 Makerere University School of Public Health, Kampala, Uganda 9 Department of International Health, Johns Hopkins University, Baltimore, MD, USA 10 Medical Research Council Laboratories, Fajara, The Gambia

Summary Routine vaccination programmes have led to substantial declines in the incidence of most of the target diseases. In these circumstances, effects beyond those on the target diseases may become evident. Several studies have suggested that certain vaccines may influence mortality in low income settings in ways that cannot be attributed to effects on target diseases. Trials of such ‘non-specific’ effects are difficult if not impossible to organise; and observational studies of them are prone to serious confounding, because those who do or do not receive vaccines are likely to differ in many ways, some of which relate to their subsequent risk of early , independent of vaccination. They are also prone to other biases, including the selective loss of vaccination records for children who die. We review these potential sources of bias and suggest what and how data may be collected to optimise the validity of such studies.

keywords vaccines, mortality, non-specific effects, study design

DTP vaccines might be associated with mortality increases Introduction (Anonymous 1981; Clemens et al. 1988; Holt et al. 1990; Routine childhood vaccination was introduced with the Koenig et al. 1990; Chen et al. 1994; Aaby et al. 1995, intention of reducing morbidity and mortality attributable 2003; Kumar et al. 2000; Kabir et al. 2003; Miller et al. to specific target diseases. It has been immensely successful 2003; Roth et al. 2006, Steenhuis et al. 2008). These in this regard, and most of these target diseases are now at observations have attracted attention to the possibility of historically low levels throughout the world. Under important vaccine effects beyond those on the target circumstances of high vaccination coverage and low target diseases (often called ‘non-specific effects’). If true, these disease frequency, there may be an opportunity for vaccine observations raise many issues, both immunological (for effects that are not specific to the target diseases to become example, they may indicate ‘biasing’ of the immune system evident. Indeed, several studies suggest that certain routine in ways not previously appreciated) and programmatic vaccines may influence child mortality in ways which (they may suggest ways in which current vaccines or cannot be ascribed to the presence or absence of the target vaccine schedules could be improved). diseases. For example, and BCG vaccines may The assessment of such ‘non-specific’ vaccine effects is reduce mortality to a larger degree than expected due to not straight-forward. Although the most rigorous way to prevention of measles and , and under test them would be through randomised trials, such an some circumstances (e.g. the near absence of pertussis) approach is not always possible. For example, where a vaccine is recommended both nationally and internation- *The members of Working Group on Non-specific Effects of ally, and current low levels of the target disease are Vaccines are given in the Appendix. dependent upon maintaining high vaccine coverage, it is

ª 2009 Blackwell Publishing Ltd 969 Tropical Medicine and International Health volume 14 no 9 pp 969–976 september 2009

P. E. M. Fine et al. Data collection in observational studies of non-specific effects of vaccines

likely to be considered unethical to randomise children to the context of vaccine studies, it is important that child receive no vaccine. For this reason, most of the studies on identifiers be assigned and recorded at the time a birth is these effects have been observational. registered in the HDSS, whether the birth occurs in a health Observational studies of such vaccine effects are difficult facility or at home. A vaccination card should be allocated to design and to interpret, being prone to serious biases. In or examined on this occasion and all vaccinations received this study, we outline some of the methodological issues should be recorded in detail, including a record if no surrounding data collection for observational studies on vaccinations have yet been given. Study identity number non-specific vaccine effects, with particular emphasis on (ID) should be recorded on the vaccination card in a way studies with mortality outcomes. that will make it clear on subsequent follow-up if the initial card has been lost and replaced. Some HDSSs include the registration of pregnancies in their routine surveillance. In The value of health and demographic surveillance such circumstances a study ID may be pre-assigned to the systems foetus and noted on the pregnancy card as this may make it While studies of non-specific vaccine effects on morbidity easier to identify the children after delivery in a maternity [e.g. the possible association of vaccination with allergic ward in which vaccines are administered. conditions (Steenhuis et al. 2008)] may be conducted in high income countries with well-developed health infor- Vaccination data mation systems, most of the mortality-studies have been carried out in low income countries. This is in part for Vaccination data collected in HDSSs may be complete and historical reasons (several key studies were carried out in correct, or they may be wrongly recorded, or missing, in West Africa), but it may be that the high prevalence and whole or in part. This will depend upon the method and variety of infectious diseases in low income settings make care with which they are gathered. There are five general them particularly prone to multiple effects of vaccines. ways in which vaccination data can be collected: Furthermore, the high background mortality risks in such • Through continuous recording by dedicated project settings means that small proportional increases are more staff members at the point of vaccination. This easily seen than they would be against a low background. method is ideal but is the most labour and resource As such countries commonly lack comprehensive data on intensive. Its feasibility should be evaluated by inves- vaccinations and mortality, these studies have typically tigators conducting studies of vaccine effects. There been conducted within health and demographic surveil- are issues of ensuring the correct identification of each lance systems (HDSSs) in which defined populations are child, but these may be minimized by disciplined use followed longitudinally for births, deaths and migrations of vaccination cards with an ID and by asking and which may include collection of information on appropriate questions (name, birth date, residence, vaccination status. Much of the following discussion thus etc.) of the adult accompanying the child (typically the relates to the data collection possibilities within HDSSs. mother). It should be recognised that an illiterate The ideal would be to collect complete data on both mother having several children with vaccination cards vaccination and mortality, including precise dates of may not have brought the correct card for the child vaccination and death, along with data on causes of death being presented for vaccination. and relevant confounding variables, for all individuals in This approach requires placing project staff in all the population under surveillance. However, while mor- vaccination clinics serving the study population. In tality data are collected in all HDSSs, cause of death addition, there need to be methods for capturing information is generally poor. Similarly, while vaccination information on vaccinations performed outside the data may be collected in some HDSSs, they are often study area. This information may (and should) be recorded in ways which are not well suited for analysis in recorded on a child’s vaccination card, but is sometimes relation to subsequent mortality. on separate pieces of paper. Mothers should thus be asked routinely if the child has been vaccinated Methodological issues in the conduct of observational elsewhere, and if these vaccinations were recorded. studies of vaccines • Through periodic review of vaccination centre records. The quality of routine records in busy Identification vaccination clinics is often poor, and it may be Each HDSS develops methods for identifying individual difficult to ensure accurate identification of children residents that are appropriate for its own circumstances. In (Luman et al. 2008). In some HDSSs measures may be

970 ª 2009 Blackwell Publishing Ltd Tropical Medicine and International Health volume 14 no 9 pp 969–976 september 2009

P. E. M. Fine et al. Data collection in observational studies of non-specific effects of vaccines

taken to improve such record keeping, for example by should be recorded in full at each visit. The date when encouraging use of project IDs in their registers. Such the card is seen is critical, as this is the date which is records may be used to validate or complement data used in landmark analysis [see companion paper on vaccinations collected in household visits. (Farrington et al. 2009)]. Figure 1 gives an example of • By transcribing from patient-held vaccination cards a data collection form for vaccinations which includes on periodic visits. This is the method used in most all the critical information. HDSSs. It raises several problems. Most obviously, it • By interviewing the parent or guardian (usually depends upon these records being completely and mother). This approach is necessary if the vaccina- correctly filled in at the time of vaccination, and then tion card is not seen at a household visit. Even if the card preserved by parents. Several studies have found that is seen, it may be appropriate to ask if vaccinations have such cards often contain information which is obvi- been received that were not recorded on the card, such ously wrong, for example inconsistent dates (Jahn as might occur if the mother forgot to take the card to a et al. 2008). The extent of these problems will be vaccination clinic or the child was vaccinated some- reduced if follow-up visits are frequent, with careful where outside the study area. Parental recall of detailed review of the cards. Some cards will be lost. In some vaccination events is likely to be faulty (Valadez & populations, lost cards will be replaced – with or Weld 1992; Simpson et al. 2001; Czaja et al. 2005), without transcription of previous vaccination data. It and should be used with great caution, if at all, in is important to record if this has occurred. analyses. In some countries different vaccines are A particularly important problem arises when infants administered in different parts of the body, for example, or children die, as in some cultures it is customary for BCG in the upper arm, DTP ⁄ Pentavalent vaccine in the parents to destroy the health records (on which thigh, and in the shoulder. This may vaccine history is recorded) of deceased children. In facilitate the mother’s correct identification of different many HDSSs a home visit is organised after the death vaccines (Aaby et al. 1998). of a child, at which time a verbal autopsy may be • By examining the child for a BCG scar. BCG is the completed to try to ascertain the cause of death. only routinely administered vaccine that leaves a scar, Depending upon the timing of this visit, it may be but correctly identifying such scars is more difficult possible to recover the vaccination card, but in some than is widely appreciated (Floyd et al. 2000; Pereira circumstances this will not be possible. Such selective et al. 2001; Roth et al. 2005). Examination for a BCG loss of vaccination data is a potential source of scar should be done by experienced staff who are important bias in analyses of the relationship between blind to whatever might be recorded on the vaccina- vaccination and mortality, as vaccinations which take tion card about BCG vaccination. Moreover, the place between survey rounds will be known only for examiner should record his or her assessment as yes, survivors, but not for those who have died. In general, no or ‘possible’. The validity of BCG scar assessment this will result in fewer vaccinations being recorded as an indicator of BCG vaccination appears to differ for children who die (Jensen et al. 2005). The mag- between populations. Studies in Brazil have shown nitude of this bias increases with increasing duration high sensitivity of scar reading (Pereira et al. 2001) between survey visits (Jensen et al. 2007), and but work in Malawi has shown that 20–25% of BCG increasing the frequency of visits ensures more com- vaccinations in early infancy do not leave lasting plete vaccination data. Another measure to improve recognisable scars (Floyd et al. 2000). The proportion data quality is to encourage card retention by parents. of children with apparent scars but no recorded BCG In some circumstances it may be appropriate to offer vaccination may be used as an indicator of unrecorded some incentive to parents for card retention. Analytic vaccinations, sometimes attributable to oversight on methods to reduce this bias have been discussed (Aaby the part of the vaccinator, to infants being vaccinated et al. 2007; Jensen et al. 2007) and are summarised in at another health centre, to administration before a a companion paper (Farrington et al. 2009). vaccination card is allocated, or to a first vaccination It is important to record certain items of information card having been lost (Jahn et al. 2008). at the time of survey visits to a study child. In With respect to the recording of vaccination data, it is of particular, record should be made of whether a the utmost importance that data on all deceased children vaccination card is actually seen at each visit, or the be as complete and correct as possible. Efforts should thus reason for its not being seen (lost or not available), be made to use all possible data sources for such children, and the informant’s identity. All vaccine-related including checking through clinic records. events that have occurred since the last card review

ª 2009 Blackwell Publishing Ltd 971 Tropical Medicine and International Health volume 14 no 9 pp 969–976 september 2009

P. E. M. Fine et al. Data collection in observational studies of non-specific effects of vaccines

Date of visit __/__/__ Infant/Child name and identifier …. Location of visit Informant (relation with child): Seen vaccination card or other written record ? Yes BCG Yes /_/ Date __/__/__ No /_/ OPV at birth Yes /_/ Date __/__/__ No /_/ DTP1 Yes /_/ Date __/__/__ No /_/ DTP2 Yes /_/ Date __/__/__ No /_/ /_/ DTP3 Yes /_/ Date __/__/__ No /_/ OPV1 Yes /_/ Date __/__/__ No /_/ OPV2 Yes /_/ Date __/__/__ No /_/ OPV3 Yes /_/ Date __/__/__ No /_/ Measles vaccine 1 Yes /_/ Date __/__/__ No /_/ Measles vaccine 2 Yes /_/ Date __/__/__ No /_/ DTP4/booster Yes /_/ Date __/__/__ No /_/ OPV4/booster Yes /_/ Date __/__/__ No /_/ Vitamin A Yes /_/ Date __/__/__ No /_/ Vitamin A Yes /_/ Date __/__/__ No /_/ Other vaccines Yes /_/ Which Date __/__/__ Other vaccines Yes /_/ Which Date __/__/__

No Claims to have card but not available /_/ Has lost card /_/ Has no card but received vaccinations /_/ Has no card, received no vaccinations /_/

Has card been Figure 1 Example of vaccination data replaced? collection form. Such a form should be completed in toto the first time the child is Comments seen. At subsequent survey rounds the form should be printed out including all the information which had already been collected, allowing a check of previously Note: If vitamin A distribution is very regular it may be necessary to have more lines for collected data and only new information this category. needs to be recorded.

health centre are associated with low or delayed vaccine Cofactors, confounders and selection bias uptake, and have also been associated with high mortality Infants and children who receive vaccines are likely to be in many populations (Jahn et al. 2008). Conversely, the different in many ways from those who do not, and this association between privileged socio-economic status and poses an obvious problem for observational studies. high vaccination coverage should reduce mortality rate A large literature on vaccine uptake patterns in different ratios comparing vaccinees to non-vaccinees. Careful populations has identified factors associated with being attention should be paid to the collection of data on such vaccinated or not (Cutts et al. 1989, 1991; Fine & Chen factors, and their use in the adjustment of analyses. 1992; Kristensen et al. 2000; Smith et al. 2001, 2005; There are, in addition, factors relating to the child’s Breiman et al. 2004; Vaugelade et al. 2004; Moulton et al. health, which influence vaccination status. For example, 2005; Chan et al. 2007; Jahn et al. 2008). Many of these children may be less likely to receive a vaccine because they factors, particularly those related to socio-economic status, are acutely sick or considered too weak or chronically ill to are also associated with mortality risks, and are hence be vaccinated. Such children may be more likely to die and potential confounders when studying associations between thus any survival advantage attributable to vaccination may vaccination and mortality. For example poverty, low be over-estimated. In many societies, low birth-weight parental education, orphan status and distance from a children are vaccinated later for reasons of both maternal

972 ª 2009 Blackwell Publishing Ltd Tropical Medicine and International Health volume 14 no 9 pp 969–976 september 2009

P. E. M. Fine et al. Data collection in observational studies of non-specific effects of vaccines

and public policy (Roth et al. 2004). As such children Mortality by cause experience relatively high mortality their inclusion in a study The evidence for non-specific effects of vaccines on will tend to inflate the mortality rate in unvaccinated mortality, whether positive or negative, is still contro- children. These effects can be subtle and strong. For example versial and the biological mechanisms for such effects are a large cohort analysis of the relation between pertussis not understood. Clearly, additional insight would be vaccines and sudden infant death syndrome (SIDS) suggested gained if the disparities in mortality were related to that the vaccine was highly protective. However, this result particular causes of death. Inferring cause of death of was almost certainly attributable to the fact that children at infants and young children is often difficult, particularly high short-term risk of SIDS had prodromal symptoms or in high mortality settings, but the importance of such other signs which had been treated as contraindications by information provides a strong motivation to devote vaccine providers (Fine & Chen 1992). The opposite bias is resources to careful verbal autopsies and medical record also possible as in some circumstances sick children may be reviews, to obtain clues to any particular causes of death vaccinated preferentially, because they come more fre- which may be more or less frequent as a function of quently to health centres. This may bias towards vaccination vaccination status. being associated with higher mortality. It is important that these potential biases are addressed in observational studies of the associations between vac- Additional aspects of observational studies of vaccines cines and mortality. Collection of data on socio-economic Campaigns indicators, distance from vaccination centres, nutritional status, birth weight and previous hospitalisations ⁄ consul- Vaccination campaigns have been carried out in many tations may enable adjustment to be made (though whether countries in recent years, in particular for measles and residual biases remain will generally be uncertain, as there polio. In many populations these doses are not routinely may be unmeasured confounding factors). Any disparity in recorded. Investigators of potential vaccine effects thus potential confounding variables between vaccinated and need to know the history of any campaigns in their study unvaccinated children should be discussed in publications. populations. In a HDSS it may be possible to collect reasonably reliable information on vaccines given to individuals by interview at the round following the Defining unvaccinated groups campaign. Information on campaigns with vitamin A Absence of evidence for vaccination is not evidence for supplementation should also be documented as some absence of vaccination. It is important to record vaccina- studies have suggested an interaction between this vitamin tion information in detail at each opportunity, e.g. at each and some vaccines (Benn et al. 2003, 2009; Moulton et al. household visit. Even if a child is brought to a vaccination 2005). centre he or she may not be vaccinated (there may be a shortage of vaccine, or the staff may be reluctant to open a Natural experiments new multi-dose vial); or a vaccination may not be recorded. The ‘unvaccinated’ group may be biased because it Natural experiments may provide an opportunity to disproportionately includes children with malformations, avoid the problems of selection bias that can be chronic diseases or who were absent at a survey round. For associated with routine vaccination programmes. There example, in a study from Malawi (Aaby et al. 2006), the may be periods when certain vaccines are missing due to association between vaccination and the risk of death was supply shortfall, or when immunization activities are very different depending on whether absent ‘unvaccinated’ otherwise interrupted, making access to one or more children were included or whether only unvaccinated vaccines determined more by logistic factors than children seen in the community were included. selection (Aaby et al. 2004). Although it may be difficult Vaccination data may be preferentially lost for children to conduct specific studies in these circumstances, it may who die, and if they are incorrectly classified as ‘unvacci- be easier in situations where a vaccine has been abruptly nated’ serious bias may arise (Farrington et al. 2009). In introduced or removed and total community mortality in some circumstances bias may be reduced by excluding relevant age groups can be compared before and after children with no recorded vaccinations from analysis. the change. Such comparisons may be relatively unbiased This approach was taken in a recent analysis from Cebu as long as the change in vaccine policy has not been in The Philippines (Chan et al. 2007), in which the effect accompanied by other simultaneous changes in of DTP was assessed only among children who had population circumstances (Desgrees du Lou et al. 1995; received BCG. Garly et al. 2004).

ª 2009 Blackwell Publishing Ltd 973 Tropical Medicine and International Health volume 14 no 9 pp 969–976 september 2009

P. E. M. Fine et al. Data collection in observational studies of non-specific effects of vaccines

implication. It is essential that care is taken at each step of Hospital in-patient survival and vaccination status investigations and full details are provided in publications, Another approach that has been used to investigate the in order that the scientific and public health communities relationship between vaccines and mortality is to relate can interpret critically and respond appropriately. admission-diagnosis-specific in-patient survival to vacci- nation status recorded at the time of admission to hospital Acknowledgements (Aaby et al. 2004; Veirum et al. 2005). Such studies are open to serious confounding related both to the reasons for This paper is the product of a workshop that was held in differential vaccination status of the patients and the London in April 2008. We are grateful to the Novo factors associated with their being brought to hospital. Nordisk Foundation and the Danish National Research Unless it is possible to collect a complete vaccination Foundation for their financial support and encouragement history from a card along with detailed data on potential and to the World Health Organisation for supporting the confounders (socio-economic, etc) such studies are likely to participation of a number of delegates. be misleading. In recognition of these difficulties, some investigators have concentrated upon male:female mortal- References ity rate ratios in such contexts, under the assumption that these are less likely to be biased than vaccinated:unvacci- Aaby P, Samb B, Simondon F, Seck AM, Knudsen K & Whittle H nated rate ratios (Garly et al. 2004; Veirum et al. 2005). (1995) Non-specific beneficial effect of measles immunisation: The potential utility of this approach is dependent on the analysis of mortality studies from developing countries. British assumption of differential vaccine effects according to Medical Journal 311, 481–485. gender and it requires further critical appraisal. Aaby P, Martins C, Bale C & Lisse I (1998) Assessing measles vaccination coverage by maternal recall in Guinea-Bissau. Lancet 352, 1229. Discussion Aaby P, Jensen H, Samb B et al. (2003) Differences in female-male mortality after high-titre measles vaccine and association with Observational studies of non-specific vaccine effects pose subsequent vaccination with diphtheria-tetanus-pertussis and particularly difficult methodological challenges, given the inactivated poliovirus: reanalysis of West African studies. Lan- obvious non-comparability of groups of individuals who cet 361, 2183–2188. have and who have not received one vaccine or another. Aaby P, Rodrigues A, Biai S et al. (2004) Oral polio vaccination Nevertheless, there are clear reasons to attempt such and low case fatality at the paediatric ward in Bissau, Guinea- studies, given the general need for monitoring vaccine Bissau. Vaccine 22, 3014–3017. effects, the importance of the questions at issue, the Aaby P, Vessari H, Nielsen J et al. (2006) Sex differential effects of evidence to date in favour of a variety of non-target disease routine immunizations and childhood survival in rural Malawi. Pediatric Infectious Disease Journal 25, 721–727. effects of various vaccines, and the impossibility of Aaby P, Benn CS, Nielsen J, Lisse IM, Rodrigues A & Jensen H carrying out formal randomised trials for every circum- (2007) DTP vaccination and child survival in observational stance. studies with incomplete vaccination data. Tropical Medicine & In this study, we have reviewed issues related to International Health 12, 15–24. collection of data for such studies, with emphasis upon Anonymous (1981) Influence of measles vaccination on survival morbidity outcomes and the use of health and demographic pattern of 7-35-month-old children in Kasongo, Zaire. The surveillance systems (HDSSs). It is the reports of mortality Kasongo Project Team. Lancet 1, 764–767. effects, beyond those explicable to changes in target disease Benn CS, Bale C, Sommerfelt H, Friis H & Aaby P (2003) incidence, which have raised the greatest concern to date, Hypothesis: Vitamin A supplementation and childhood mor- for obvious reasons, and these deserve critical attention in tality: amplification of the non-specific effects of vaccines? International Journal of Epidemiology 32, 822–828. the future. HDSSs provide appropriate settings for such Benn CS, Martins C, Rodrigues A et al. (2009) The effect of research, and it is hoped that this review will encourage vitamin A supplementation administered with missing vaccines these sites to devote attention and resources to this during national immunization days in Guinea-Bissau. Interna- important subject. The fact that vaccines and vaccination tional Journal of Epidemiology 38, 304–311. programmes continue to evolve means that the potential Breiman RF, Streatfield PK, Phelan M, Shifa N, Rashid M & for complex effects of vaccines will also evolve in the years Yunus M (2004) Effect of infant immunisation on childhood to come. mortality in rural Bangladesh: analysis of health and demo- The issue of non-specific effects of vaccines is compli- graphic surveillance data. Lancet 364, 2204–2211. cated in many ways, in terms of data collection, analysis, Chan GJ, Moulton LH, Becker S, Munoz A & Black RE (2007) interpretation, biological mechanism and programmatic Non-specific effects of diphtheria tetanus pertussis vaccination

974 ª 2009 Blackwell Publishing Ltd Tropical Medicine and International Health volume 14 no 9 pp 969–976 september 2009

P. E. M. Fine et al. Data collection in observational studies of non-specific effects of vaccines

on child mortality in Cebu, The Philippines. International Koenig MA, Khan MA, Wojtyniak B et al. (1990) Impact of Journal of Epidemiology 36, 1022–1029. measles vaccination on childhood mortality in rural Bangladesh. Chen RT, Weierbach R, Bisoffi Z et al. (1994) A ‘post-honeymoon Bulletin of the World Health Organization 68, 441–447. period’ measles outbreak in Muyinga sector, Burundi. Interna- Kristensen I, Aaby P & Jensen H (2000) Routine vaccinations and tional Journal of Epidemiology 23, 185–193. child survival: follow up study in Guinea-Bissau, West Africa. Clemens JD, Stanton BF, Chakraborty J et al. (1988) Measles British Medical Journal 321, 1435–1438. vaccination and childhood mortality in rural Bangladesh. Kumar G, Anand K, Kant S & Kapoor SK (2000) Scale for iden- American Journal of Epidemiology 128, 1330–1339. tification of ‘‘at risk’’ families for underfive deaths. Indian Cutts FT, Rodrigues LC, Colombo S & Bennett S (1989) Evalua- Journal of Pediatrics 67, 411–417. tion of factors influencing vaccine uptake in Mozambique. Luman ET, Sablan M, Stokley S, McCauley MM & Shaw KM International Journal of Epidemiology 18, 427–433. (2008) Impact of methodological ‘‘shortcuts’’ in conducting Cutts FT, Diallo S, Zell ER & Rhodes P (1991) Determinants of public health surveys: results from a vaccination coverage sur- vaccination in an urban population in Conakry, Guinea. Inter- vey. BMC Public Health 8, 99. national Journal of Epidemiology 20, 1099–1106. Miller E, Andrews N, Waight P & Taylor B (2003) Bacterial infec- Czaja C, Crossette L & Metlay JP (2005) Accuracy of adult tions, immune overload, and MMR vaccine. Measles, mumps, reported pneumococcal vaccination status of children. Annals of and rubella. Archives of Disease in Childhood 88, 222–223. Epidemiology 15, 253–256. Moulton LH, Rahmathullah L, Halsey NA, Thulasiraj RD, Katz J Desgrees du Lou A, Pison G & Aaby P (1995) Role of immuni- & Tielsch JM (2005) Evaluation of non-specific effects of infant zations in the recent decline in childhood mortality and the immunizations on early infant mortality in a southern Indian changes in the female ⁄ male mortality ratio in rural Senegal. population. Tropical Medicine & International Health 10, 947– American Journal of Epidemiology 142, 643–652. 955. Farrington CP, Firth M, Moulton LH et al. (2009) Epidemiolog- Pereira SM, Dourado I, Barreto ML et al. (2001) Sensitivity and ical studies of the non-specific effects of vaccines: II – method- specificity of BCG scar reading in Brazil. International Journal ological issues in the design and analysis of cohort studies. of Tuberculosis & Lung Disease 5, 1067–1070. Tropical Medicine & International Health in press. Roth A, Jensen H, Garly ML et al. (2004) Low birth weight Fine PE & Chen RT (1992) Confounding in studies of adverse infants and Calmette-Guerin bacillus vaccination at birth: reactions to vaccines. American Journal of Epidemiology 136, community study from Guinea-Bissau. Pediatric Infectious 121–135. Disease Journal 23, 544–550. Floyd S, Ponnighaus JM, Bliss L et al. (2000) BCG scars in Roth A, Sodemann M, Jensen H et al. (2005) Vaccination tech- northern Malawi: sensitivity and repeatability of scar reading, nique, PPD reaction and BCG scarring in a cohort of children and factors affecting scar size. International Journal of Tuber- born in Guinea-Bissau 2000–2002. Vaccine 23, 3991–3998. culosis & Lung Disease 4, 1133–1142. Roth A, Garly ML, Jensen H, Nielsen J & Aaby P (2006) Bacillus Garly ML, Jensen H, Martins CL et al. (2004) Hepatitis B vacci- Calmette-Guerin vaccination and infant mortality. Expert nation associated with higher female than male mortality in Review of Vaccines 5, 277–293. Guinea-Bissau: an observational study. Pediatric Infectious Simpson DM, Ezzati-Rice TM & Zell ER (2001) Forty years and Disease Journal 23, 1086–1092. four surveys: how does our measuring measure up? American Holt EA, Boulos R, Halsey NA, Boulos LM & Boulos C (1990) Journal of Preventive Medicine 20, 6–14. Childhood survival in Haiti: protective effect of measles vacci- Smith PJ, Battaglia MP, Huggins VJ et al. (2001) Overview of the nation. Pediatrics 85, 188–194. sampling design and statistical methods used in the National Jahn A, Floyd S, Mwinuka V et al. (2008) Ascertainment of Immunization Survey. American Journal of Preventive Medicine childhood vaccination histories in northern Malawi. Tropical 20, 17–24. Medicine & International Health 13, 129–138. Smith PJ, Hoaglin DC, Battaglia MP, Khare M, Barker LE & Jensen H, Benn CS & Aaby P (2005) DTP in low income countries: Centers for Disease Control and Prevention (2005) Statistical improved child survival or survival bias? British Medical Journal methodology of the National Immunization Survey, 1994–2002. 330, 845–846. Vital & Health Statistics – Series 2: Data Evaluation & Methods Jensen H, Benn CS, Lisse IM, Rodrigues A, Andersen PK & Aaby P Research 138, 1–55. (2007) Survival bias in observational studies of the impact of Steenhuis TJ, van Aalderen WM, Bloksma N et al. (2008) Bacille- routine immunizations on childhood survival.[erratum appears Calmette-Guerin vaccination and the development of allergic in Trop Med Int Health. 2007 Jul;12(7):902]. Tropical Medicine disease in children: a randomized, prospective, single-blind & International Health 12, 5–14. study. Clinical & Experimental Allergy 38, 79–85. Kabir Z, Long J, Reddaiah VP, Kevany J & Kapoor SK (2003) Valadez JJ & Weld LH (1992) Maternal recall error of child Non-specific effect of measles vaccination on overall child vaccination status in a developing nation. American Journal of mortality in an area of rural India with high vaccination cov- Public Health 82, 120–122. erage: a population-based case-control study. Bulletin of the Vaugelade J, Pinchinat S, Guiella G, Elguero E & Simondon F World Health Organization 81, 244–250. (2004) Non-specific effects of vaccination on child survival:

ª 2009 Blackwell Publishing Ltd 975 Tropical Medicine and International Health volume 14 no 9 pp 969–976 september 2009

P. E. M. Fine et al. Data collection in observational studies of non-specific effects of vaccines

prospective cohort study in Burkina Faso. British Medical Oslo, Norway; Honorati Masanja, Ifakara Health Journal 329, 1309. Research and Development Centre, Ifakara, Tanzania; Veirum JE, Sodemann M, Biai S et al. (2005) Routine vaccinations Puneet Misra, Center for Community Medicine, All India associated with divergent effects on female and male mortality Institute of Medical Sciences, New Delhi, India; Law- at the paediatric ward in Bissau, Guinea-Bissau. Vaccine 23, rence H Moulton, Division of International Health, 1197–1204. Johns Hopkins University, Baltimore, USA; Kim Mul- holland, London School of Hygiene & Tropical Medi- Appendix cine, London, UK; Martin Mutua, African Population & Health Research Center, Nairobi, Kenya; Jens Nielsen, Members of Working Group on Non-specific Effects of Bandim Health Project, Statens Seruminstitut, Copenha- Vaccines: Peter Aaby, Bandim Health Project, Bissau, gen, Denmark; Hanna Nohynek, National Public Health Guinea-Bissau and Danish Epidemiology Science Centre, Institute Finland and Department of International Statens Seruminstitut, Copenhagen, Denmark; Martin Health, University of Tampere, Finland; Seth Owusu- Adjuik, Navrongo Health Research Centre, Upper East Agyei, Kintampo Health Research Centre, Kintampo, Region, Ghana; Gert Almind, Novo Nordisk Foundation, Ghana Henrik Ravn, Bandim Health Project, Statens Copenhagen, Denmark; Per Kragh Andersen, Department Seruminstitut, Copenhagen, Denmark; Abdur Razzaque, of Biostatistics, University of Copenhagen, Copenhagen, Matlab Health and Demographic Surveillance Site, Denmark; Ashish Bavdekar, Vadu Rural Health Program, ICDDR,B, Dhaka, Bangladesh; Laura Rodrigues, London KEM Hospital, Pune, India; Christine S Benn, Bandim School of Hygiene & Tropical Medicine, London, UK; Health Project, Statens Seruminstitut, Copenhagen, Den- Amabelia Rodrigues, Bandim Health Project, Bissau, mark; Stephen Evans, London School of Hygiene & Guinea-Bissau; Anthony Scott, Kenya Medical Research Tropical Medicine, London, UK; C Paddy Farrington, Institute Centre for Geographic Medicine, Kilifi, Kenya Department of Mathematics and Statistics, The Open and Nuffield Department of Medicine, Oxford Univer- University, Milton Keynes, UK; Paul EM Fine, London sity, Oxford, UK; Frank Shann, Royal Childrens Hospi- School of Hygiene & Tropical Medicine, London, UK; tal, University of Melbourne, Melbourne, Australia; Ali Martin J Firth, Telethon Institute for Child Health Sie, Centre de Recherche en Sante de Nouna, Burkina Research and Centre for Child Health Research, The Faso; Kirsten B Simondon, Epidemiology and Prevention University of Western Australia, Australia; Katie L Flan- Research Unit, Institut de Recherche pour le De´velopp- agan, MRC Laboratories, Fajara, The Gambia; Anneke ement, Montpellier, France; Peter G Smith, London Hesseling, Desmond Tutu Tuberculosis Centre, Stel- School of Hygiene & Tropical Medicine, London, UK; lenbosch University, South Africa; Siddhivinayak Hirve, Paul Welaga, Navrongo Health Centre, Upper East Vadu Rural Health Program, KEM Hospital, Pune, India; Region, Ghana; Thomas N Williams, Kenya Medical Jasseh Momodou, MRC Laboratories, Fajara, The Gam- Research Institute Centre for Geographic Medicine, bia; Karin Ka¨llander, Division of International Health, Kilifi, Kenya and Nuffield Department of Medicine, Karolinska Institute, Stockholm, Sweden and School of Oxford University, Oxford, UK; Kapil Yadav, Ballabgarh Public Health, Makerere University, Kampala, Uganda; Health and Demographic Surveillance Site, All Institute Anand Krishnan, Ballabgarth HDSS, CCM, AIIMS, New of Medical Sciences, Haryana, India. Delhi, India; Jan Maehlen, Ullevaal University Hospital,

Corresponding Author Paul E. M. Fine, London School of Hygiene & Tropical Medicine, Keppel Street, London WC1E 7HT, UK. Tel.:+44 20 927 2219; Fax: +44 20 7636 8739; E-mail: paul.fi[email protected]

976 ª 2009 Blackwell Publishing Ltd