This article has Open Peer Review reports available.
Agreement between physicians and the InterVA-4 model in assigning causes ofdeath: the role of recall period and characteristics specific to the deceasedand the respondent
© Tadesse; licensee BioMed Central Ltd. 2013
Received: 13 April 2013
Accepted: 14 October 2013
Published: 6 November 2013
In the absence of routine death registration, the InterVA model is a newmethodology being used as a physician alternative method to interpret verbalautopsy (VA) data in resource-poor settings. However, various studiesindicate that there are significant discrepancies between the two approachesin assigning causes of deaths. This study evaluated the role of recallperiod and characteristics that were specific to the deceased and therespondent in affecting the level of agreement between the approaches.
A population-based cross-sectional study was conducted from March to April,2012. All adults aged ≥14 years and died between 01 January,2010, and 15 February, 2012, were included in the study. Data were collectedby using a pre-tested and modified WHO designed verbal autopsyquestionnaire. The verbal autopsy interviews were reviewed by the InterVA-4model and the physicians. Cohen’s kappa statistic with 95% CI wasapplied to compare the strength of the agreement between the model and thephysician review.
A total of 408 VA interviews were successfully completed and reviewed by theInterVA model and the physicians. Both approaches showed an overallagreement in 294 (72.1%) of the cases [kappa = 0.48, 95% CI:0.42 - 0.60]. The level of agreement between the approaches was low[kappa ≤0.40] when the deceased was female, 50 and above years old,single, illiterate, rural dweller, belonged to a family of 1–4 peopleliving together, and died at home. This was also true when the recall periodwas ≤1 year, and the respondent was a relative other thanparent/marital partner, lived with the deceased, and had medicalinformation.
This study identified important variables affecting the strength of agreementbetween the InterVA-4 model and the physician in assigning causes of death.The results are believed to significantly contribute to the process ofidentifying the actual underlying causes of deaths in the population, andmay thus serve to promote informed health policy decisions in resource-poorsettings.
Developing countries generally lack consistent, timely, and reliable information onlevels of cause specific mortality fractions (CSMFs) in their populations [1, 2]. Verbal autopsy (VA) is a useful tool in such settings to establish theprobable cause of death (COD) by interviewing a close caregiver or anyone who canprovide witness to the death event . VA data are often reviewed by physicians in order to assign the probableCOD. But in addition to being time and energy consuming, the method is likely toproduce inconsistent results [4–16]. Different alternative methods to the physician review process forinterpreting VA data have remained of limited use [17–19]. However, the use of the InterVA model to interpret VA data has just beenexplored to have the advantage of achieving the maximum spatial and temporalconsistency in interpreting VA data. Moreover, it requires less time and laborresources, especially in comparison with the physician review method [20–22]. Also, it is freely available in the public domain, making it ideal forresource-constrained settings .
Various studies have been conducted to compare the performance of the InterVA modelas a physician alternative method to interpret VA data [12, 20, 24, 25]. However, the results still show some discrepancies in comparison to thephysician review. Moreover, the role of recall period and characteristics specificto the deceased and the respondent in affecting the level of agreement between thetwo approaches have not been assessed. Therefore, this study is designed to evaluatethe role of recall period and characteristics specific to the deceased and therespondent in affecting the level of agreement between the InterVA model and thephysician. The study results are believed to significantly contribute to the processof identifying the actual underlying CODs in the population, and may thus serve topromote informed health policy decisions in resource-poor settings.
A population-based cross-sectional study was carried out from March to April, 2012,in Dabat Health and Demographic Surveillance System site (HDSSs), hosted by theUniversity of Gondar. The site is located in a district known as Dabat, northernEthiopia, which has an estimated population of 46,165 living in 7 rural and 3 urban“kebeles” (the smallest administrative units in Ethiopia). The localcommunities largely depend on subsistence agriculture. Information on vital events,like birth, death, and migration are collected quarterly .
Study population and data collection
All adults aged ≥14 years and died between 01 January, 2010, and 15February, 2012, in the area were included in the study. This period waspreferred in order to obtain an adequate number of deaths without a markedrecall bias. It is believed that adult deaths were remembered very well.
Pre-tested and modified WHO and INDEPTH [27, 28] designed VA questionnaire was used to collect the data. The VAquestionnaire included open narrative, medical histories, and closed questions.The narrative section was used to record free explanations of the circumstancesof deaths; the medical history sections were used to extract data from medicalcertificates, and the closed section dealt with specific signs, symptoms, andconditions leading to death. Three trained supervisors and nine data collectorswho had rich experience in the job participated in the data collectionprocesses. After obtaining informed written consents, the data collectorsinterviewed a close relative, a friend, or neighbors of the deceased person whowitnessed the death.
The VA questionnaire was translated into “Amharic” (the locallanguage) and back to English to maintain the consistency of the questions. Thetraining of the data collectors and supervisors emphasized issues, such as theselection of eligible respondents, approaching grieving respondents, time ofinterviews, and compiling narrative responses (ensuring that duration,frequency, severity, and sequence of symptoms were mentioned). The principalinvestigator and the supervisors coordinated the interview process, madespot-checks, and reviewed the completed questionnaire on daily bases to ensurethe completeness and consistency of the data collected .They also conductedrandom quality checks by re-interviewing about 10% of the respondents. The VAquestionnaire was pre-tested to identify potential problem areas, unanticipatedinterpretations, and cultural objections to any of the questions on 25respondents (near Dabat district) with characteristics similar to the studysubjects. Based on the pre-test results, the questionnaire was adjustedcontextually. Data entry was carried out by the principal investigator andanother independent data clerk and was then compared to check for any variationin results.
Interpretation of VA data
The InterVA-4 model and the physician reviewed the same basic data from the VAquestionnaire independently. That is, both methods utilized information in theopen narrative, medical history, and the closed-ended section to assign theprobable COD.
Two independent physicians reviewed each VA questionnaire independently to assigna single COD based on ICD-10. The ICD-10 list had unique codes for diseases,signs, symptoms, abnormal findings, complaints, social circumstances, andexternal causes of injury . The physicians met subsequently to reach consensus on cases wherethere were differences of opinion. If no physician consensus was reached afterdiscussion, the COD was regarded as indeterminate. The physicians were trainedin procedures on assigning COD and given details of the study area and studypopulation. However, they were not given any special briefings on theprobabilistic model so as not to encroach on their professional freedom. Inspite of that, however, their review process was closely monitored and that theybe not direct beneficiaries of the research output was ensured.
Interpretation of the InterVA model
The model relates a range of input indicators, such as age, sex, physical signsand symptoms, medical history, and the circumstances of death to likely CODsusing the Bayesian probabilities . The model results in up to three likely causes per case whenpossible; each associated with a quantified likelihood. To assign an estimate ofthe overall certainty for that patient, the model gives the average likelihoodfor a maximum of three CODs . In this study, a high prevalence of malaria and HIV/AIDS were usedas basic epidemiological parameters for the model as their prevalence variedfrom place to place. Data were entered into the already specified batchin.csvfile format of InterVA version 4, and a readable text output log file format waschosen to assign the possible COD responsible for the death of eachindividual.
Agreement between the InterVA model and the physician
The most probable CODs assigned by the model were considered to facilitatecomparison with the single CODs which were assigned by the physician. In a casewhere there was more than one probable CODs provided by the InterVA-4 model, thesecond and the third, if any, CODs were considered to compare the agreementbetween the model and the physician reviews. Agreement between the twoapproaches was sought at chapter heading level of ICD-10. All CODs in bothmethods were re-categorized into 9 main groups for two reasons. The first reasonwas to have meaningfully comparable COD categories between both methods. Second,it was more important that the model and the physician arrive at a broadagreement in identifying COD groups with the greatest public health importanceat population level, rather than individual level causes. The 9 main categoriesused in this study were the following: pulmonary tuberculosis, kidney diseases,liver diseases, diabetes, other infectious diseases, cardiovascular problems,maternity-related deaths, other non-communicable diseases, andinjuries/accidents.
Then deaths were aggregated case-by-case to their respective COD categories inorder to determine the CSMFs at the community level by using both the InterVAmodel and the physician review. Cohen’s kappa statistic (K) with 95%confidence interval (CI) was applied to compare the agreement between theInterVA model and the physician review. Complete agreement corresponds to a K of1 and complete disagreement to a K of 0. The strength of agreement was rated aslow for a K ≤ 0.40, moderate for a Kbetween 0.41 and 0.60, good for a K between 0.61 and 0.80, and verygood for a K > 0.80 .
The study protocol was reviewed and approved by the Institutional Ethical ReviewBoard of the University of Gondar. Then, informed written consent was obtainedfrom the study participants who were close relatives, friends, or neighbors ofthe deceased after explaining the purpose and the procedures of the study.Confidentiality was guaranteed for information collected from each studyparticipant. Study participants found sick at the time of data collection werereferred to the nearest health institution for medical treatment. There was noremuneration for family.
Finally, for the purpose of completeness findings of the previous studies on thephysician reviews of the VA data were included in this study [13, 30]. The current and the previous studies were conducted in the samestudy area, data source, and study period.
A total of 408 VA interviews were successfully completed and reviewed by both thephysicians and the InterVA model.
Out of the 408 deaths, 329 (80.6%) were successfully assigned a single cause atthe first attempt by two physicians. After holding consensus meetings, thephysicians readily assigned a single COD to 61 (15%) more cases. Therefore, onthe whole, physicians assigned a single COD to 390 (95.6%) cases. No consensuswas reached on 18(4.4%) cases which were coded as “indeterminate” bythe physicians.
Interpretation of the InterVA model
The InterVA model assigned a single COD to 347 (85.1%) cases, two CODs to 40(9.8%) cases, and three causes to 3(0.7%) cases. In 18 (4.4%) cases, the InterVAmodel assigned the COD as “indeterminate”.
Agreement between the InterVA model and the physician
Proportion of CSMFs by the physician and by the InterVA model inDabat, Ethiopia from 01 January 2010–15 February 2012
N = 408 (%)
N = 408 (%)
Other infectious diseases
Other non-communicable diseases
The role of socio-demographic characteristics of the deceased
Agreement between physicians and InterVA model CODs allocation bysocio-demographic characteristics of deceased in Dabat, Ethiopia
Kappa (95% CI)
0.46 (0.4, 0.6)
0.32 (0.2, 0.4)
Age in years
0.59 (0.5, 0.7)
0.25 (0.2, 0.4)
0.37 (0.3, 0.5)
0.36 (0.3, 0.5)
0.56 (0.5, 0.7)
0.38 (0.3, 0.5)
0.43 (0.3, 0.5)
0.36 (0.3, 0.5)
0.44 (0.3, 0.5)
0.42 (0.3, 0.5)
0.41 (0.3, 0.5)
0.37 (0.3, 0.5)
0.46 (0.4, 0.6)
0.36 (0.3, 0.5)
0.50 (0.3, 0.7)
0.54 (0.2, 0.8)
The role of recall period
Agreement between physicians and InterVA model CODs allocation byrecall period and characteristics of respondent in Dabat, Ethiopiafrom 01 January 2010–15 February 2012
Kappa (95% CI)
0.19 (0.0, 0.4)
0.32 (0.2, 0.4)
0.27 (0.2, 0.4)
0.53 (0.4, 0.6)
19-25 ½ months
0.44 (0.3, 0.5)
0.42 (0.3, 0.5)
0.33 (0.2, 0.4)
0.51 (0.3, 0.7)
Respondent live with status
0.35 (0.3, 0.5)
0.41 (0.3, 0.5)
The role of the characteristics of the respondent
When the relation of the respondent to the deceased was parent/marital partner,other relative and unrelated, the level of agreement between the physician andthe InterVA model in assigning CODs was moderate, low, andmoderate, respectively. The level of agreement was lowwhen the respondent lived with the deceased and good, otherwise. It waslow when respondents had medical information about the diseasecondition of the deceased and moderate when they didn’t,(Table 3).
In this study, the role of the socio-demographic characteristics of the deceased,recall period, and characteristics of respondents in influencing the level ofagreement between the physician and the InterVA-4 model in assigning CSMFs at thepopulation level was evaluated and found to be significant. A moderatelevel of agreement was found between the model and the physician in establishing allCODs [kappa = 0.48, 95% CI: 0.42 - 0.60]. Almost a similar finding wasobserved in a previous literature . This indicated the temporal and spatial consistency of the model forestablishing cause-specific mortalities.
The level of agreement between both approaches was low when the deceased wasfemale as compared to when the deceased was male in this population. This could beexplained by the low educational attainment combined with the poor health-seekingbehavior of females [27, 32–34] which might significantly influence correct symptom characterization oftheir illness conditions which in turn could possibly lead to wrong conclusions ofthe COD. Regarding the age of the deceased, a low level of agreement wasobserved for cases older than 50 and above as compared to younger ages. This couldbe justified by the simultaneous occurrence of multiple illness conditions withoverlapping symptomatic nature as a result of age which might significantlyinfluence the likelihood of COD assignment by both approaches. A low levelof agreement was observed when the deceased was single as compared to married. Thiscould be due to the fact that single people usually make infrequent and loose socialinteractions with the society. As a result, respondents could fail to correctlycharacterize the event responsible for the death when they are interviewed.Consequently, this might lead to confusion during COD assignment. There was alow level of agreement when the deceased was illiterate as compared toliterate. The possible explanation for this could be that illiterate people seldomunderstand and explain their illness conditions correctly to their relatives whichmight significantly contribute to the low level of agreement between thephysician and the InterVA model in assigning CSMFs at the population level. Forrural residents, the level of agreement between the physician and the InterVA modelwas low compared to urban dwellers. This could be due to the fact thatrural people in developing countries rarely seek appropriate modern medical servicesto correctly characterize their illness conditions although they suffer frommultiple illness conditions with overlapping symptoms which might be responsible forthe low level of agreement between the physician and the InterVA model. Thereason for the low level of agreement between the physician and the InterVAmodel when family size of the deceased was 1–4 people might be that the lowfrequency of social contacts made by the deceased during their illness with the fewfamily members who are already engaged in busy daily activities could directlyimpair their ability to correctly characterize the illness condition of the deceasedwhen they are interviewed. Consequently, the physician and the InterVA model couldreach an agreement on only few CODs. A low level of agreement was observedwhen the place of death was the home as compared to death at a health facility andother places. This could be explained by the fact that the majority of the people inthe study area were illiterate who used more local terms to characterize the deathevent which could lead the two approaches to reach different conclusions.
Currently, a wide range of recall periods from the time of death to the interview isused in VA. Some perform interviews as soon as possible after death while othersvisit the household of the deceased after a minimum of four weeks to allowattendants an adequate mourning period. The maximum recall period varied from sixmonths to an indefinite period. The effects of recall may differ depending on thecontext, characteristics and demographics of the deceased . Validation studies confirmed that a recall period ranging from1 month to 2 years is generally thought to be acceptable [36, 37]. This study supports this finding on the ground that the level ofagreement between the physician and the InterVA model in assigning CODs increased asthe recall period got longer and longer. This could be so because as the timebetween the interview and death increased, there would be a decrease in respondentrelated bias enabling the respondent to characterize the death event freely. Thiscould consequently improve the level of agreement between the two approaches inassigning the COD.
In this study, a low level of agreement between the physician and theInterVA model in assigning CODs was observed when the relation of the respondentwith the deceased was other relative (son, daughter, brother, sister, uncle, ant) asopposed to parent/marital partner, and unrelated. The reason for this could be thatother relatives would not spend most of their time with the deceased during theirillness and try to characterize the illness condition wrongly when they are asked toexplain the death event. This would consequently lower the level of agreementbetween the approaches in assigning the COD. Unlike this, parents/marital partnersare more likely to spend most of their time with the deceased during their illnessand as a result could explain the death event more accurately which may lead to animproved level of agreement in assigning the COD. Unrelated respondents rarelyintroduce respondent related bias when they are asked to explain the death event.This could contribute to the increased level of agreement between the approaches inassigning the COD. The reason for the low increase in the level ofagreement when respondents lived with the deceased might be due to the influence ofrespondents’ traditional understanding and stereotyped way of characterizingthe illness condition responsible for the death event. The level of agreementbetween the physician and the InterVA model in assigning CODs was observed to below when the respondents had medical information about the diseasecondition responsible for the death event. The possible explanation for this couldbe that respondents who had medical information stick only to naming thespecific-cause of death, ignoring the other indicators responsible for the deathevent when they were interviewed. This could affect the validity of the physicianand the InteVA model differently.
The possible limitation of this study could be that the influence of therespondent’s age, sex, marital, occupational, and educational status on thelevel of agreement between the two approaches was not evaluated. Contextually anddemographically sensitive VA studies should be conducted to address these gaps.
In this study, a low level of agreement was observed between the InterVA-4model and the physician in assigning CODs when the deceased was female, 50 and aboveyears old, single, illiterate, rural dweller, belonged to a family of 1–4people living together, and died at home. This was also true when the recall periodwas ≤1 year, and the respondent was a relative other than parent/maritalpartner, lived with the deceased, and had medical information. Therefore, inaddition to providing adequate training to data collectors on how to selectinterviewees and elicit the right indicators of the COD responsible for a particulardeath, VA researchers should choose the appropriate recall period in order togenerate high quality data to be used by the InterVA model. These techniquessignificantly contribute to the process of identifying the actual underlying CODs inthe population, and may thus serve to promote informed health policy decisions inresource-poor settings.
The author wishes to thank the University of Gondar for funding this study. Heacknowledges the Dabat District Health Office for logistic and administrativesupport, and data collectors for their support in making this study possible.Also, he extends his appreciation to Dr. Dagnachew Yohannes and Dr. Girma Lobefor assigning the causes of deaths for all the VA data. Finally, his deepestgratitude goes to the families in Dabat who participated in this study.
- Setel PW, Macfarlane SB, Szreter S, Mikkelsen L, Jha P, et al: A scandal of invisibility: making everyone count by counting everyone. Lancet. 2007, 370: 1569-1577. 10.1016/S0140-6736(07)61307-5. doi: 10.1016/S0140-6736(07)61307-5View ArticlePubMedGoogle Scholar
- Byass P: Who needs cause-of-death data?. PLoS Med. 2007, 4 (11): 333-10.1371/journal.pmed.0040333. doi: 10.1371/journal.pmed.0040333View ArticleGoogle Scholar
- Fottrell E: Dying to count: mortality surveillance in resource-poor settings. Glob Health Action. 2009, 2: 10-doi: 3402/gha.v2i0.1926PubMed CentralGoogle Scholar
- Fottrell E, Byass P: Verbal autopsy: methods in transition. Epidemiol Rev. 2010, 32: 38-55. 10.1093/epirev/mxq003. doi: 10.1093/epirev/mxq003View ArticlePubMedGoogle Scholar
- Coldham C, Ross D, Quigley M, Segura Z, Chandramohan D: Prospective validation of a standardized questionnaire for estimatingchildhood mortality and morbidity due to pneumonia and diarrhoea. Trop Med Int Health. 2000, 5: 134-144. 10.1046/j.1365-3156.2000.00505.x. doi: 10.1046/j. 1365–3156.2000.00505.x,View ArticlePubMedGoogle Scholar
- Soleman N, Chandramohan D, Shibuya K: Verbal autopsy: current practices and challenges. Bull World Health Organ. 2006, 84: 239-245. 10.2471/BLT.05.027003. doi: 10.2471/BLT.05.027003View ArticlePubMedPubMed CentralGoogle Scholar
- Ronsmans C, Vanneste AM, Chakraborty J, Ginneken JV: A comparison of three verbal autopsy methods to ascertain level and causes ofdeath in Matlab Bangladesh. Int J Epidemiol. 1998, 27: 660-6. 10.1093/ije/27.4.660. doi: 10.1093/ije/27.4.660,View ArticlePubMedGoogle Scholar
- Todd JE, De Francisco A, Dempsey TJO, Greenwood BM: The limitations of verbal autopsy in a malaria endemic region. Ann Trop Paediatr. 1994, 14: 31-36. PubMed: 7516132PubMedGoogle Scholar
- Vergnano S, Fottrell E, Osrin D, Lewycka S, Costello A, et al: Adaptation of a probabilistic method (InterVA) of verbal autopsy to improvethe interpretation of cause of stillbirth and neonatal death in Malawi,Nepal, and Zimbabwe. Popul Health Metr. 2011, 9: 48-10.1186/1478-7954-9-48. doi: 10.1186/1478-7954-9-48,View ArticlePubMedPubMed CentralGoogle Scholar
- Byass P, Kahn K, Fottrell E, Mee P, Collinson MA, et al: Using verbal autopsy to track epidemic dynamics: the case of HIV-relatedmortality in south Africa. Popul Health Metr. 2011, 9: 46-10.1186/1478-7954-9-46. doi: 10.1186/1478-7954-9-46,View ArticlePubMedPubMed CentralGoogle Scholar
- World Health Organization: Beyond the numbers: reviewing maternal deaths and complications to makepregnancy safer. 2004, Geneva: WHO,Google Scholar
- Byass P, Huong DL, Minh HV: A probabilistic approach to interpreting verbal autopsies: methodology andpreliminary validation in Vietnam. Scand J Public Health Suppl. 2003, 62: 32-37. doi: 10.1080/14034950310015086,View ArticlePubMedGoogle Scholar
- Tadesse S, Tadesse T: Evaluating the performance of interpreting verbal autopsy 3.2 Model forestablishing pulmonary tuberculosis as a cause of death in Ethiopia: apopulation-based cross-sectional study. BMC Public Health. 2012, 12: 1039-10.1186/1471-2458-12-1039.View ArticlePubMedPubMed CentralGoogle Scholar
- Misganaw A, Araya T, Aneneh A, Hailemariam D: Validity of verbal autopsy method to determine causes of death among adultsin the urban setting of Ethiopia. BMC Med Res Methodol. 2012, 12: 130-10.1186/1471-2288-12-130. doi: 10.1186/1471-2288-12-130,View ArticlePubMedPubMed CentralGoogle Scholar
- Byass P, Kahn K, Collinson MA, Tollman SM, Fottrell E: Moving from data on deaths to public health policy in Agincourt, southAfrica: approaches to analyzing and understanding verbal autopsyfindings. PLoS Med. 2010, 7 (8): e1000325-10.1371/journal.pmed.1000325. doi: 10.1371/journal.pmed.1000325,View ArticlePubMedPubMed CentralGoogle Scholar
- Freeman JV, Christian P, Khatry SK, Adhikari RK, Clerq SCL, et al: Evaluation of neonatal verbal autopsy using physician review versusalgorithm-based cause-of-death assignment in rural Nepal. Paediatr Perinat Epidemiol. 2005, 19: 323-331. 10.1111/j.1365-3016.2005.00652.x. doi: 10.1111/j.1365-3016.2005.00652.x,View ArticlePubMedGoogle Scholar
- Quigley MA, Chandramohan D, Rodrigues LC: Diagnostic accuracy of physician review, expert algorithms and data derivedalgorithms in adult verbal autopsies. Int J Epidemiol. 1999, 28: 1081-7. 10.1093/ije/28.6.1081. doi: 10.1093/ije/28.6.1081,View ArticlePubMedGoogle Scholar
- Boulle A, Chandramohan D, Weller P: A case study of using artificial neural networks for classifying cause ofdeath from verbal autopsy. Int J Epidemiol. 2001, 30 (515): 520-doi: 10.1093/ije/30.3.515,Google Scholar
- Quigley MA, Chandramohan D, Setel P, Binka F, Rodrigues LC: Validity of data-derived algorithms for ascertaining causes of adult death intwo African sites using verbal autopsy. Trop Med Int Health. 2000, 5 (1): 33-39. 10.1046/j.1365-3156.2000.00517.x. doi: 10.1046/j.1365-3156.2000.00517.x,View ArticlePubMedGoogle Scholar
- Fantahun M, Fottrell E, Berhane Y, Wall S, Hogberg U, et al: Assessing a new approach to verbal autopsy interpretation in a ruralEthiopian community: the InterVA model. Bull World Health Organ. 2006, 84: 204-210. 10.2471/BLT.05.028712. doi: 10.2471/BLT.05.028712,View ArticlePubMedPubMed CentralGoogle Scholar
- Reeves BC, Quigley M: A review of data derived methods for assigning cause of death from verbalautopsy data. Int J Epidemiol. 1997, 26 (5): 1080-1089. 10.1093/ije/26.5.1080. doi: 10.1093/ije/26.5.1080,View ArticlePubMedGoogle Scholar
- Byass P, Fottrell E, Huong DL, Berhane Y, Corrah T, et al: Refining a probabilistic model for interpreting verbal autopsy data. Scand J Public Health. 2006, 34: 26-31. 10.1080/14034940510032202. doi: 10.1080/14034940510032202View ArticlePubMedPubMed CentralGoogle Scholar
- InterVA 3.2 Model. Available: http://www.interva.net. Accessed 2012 February 12
- Byass P, Fottrell E, Witten KH, Bhattacharya S, Fitzmaurice AE, et al: Revealing the burden of maternal mortality: a probabilistic model fordetermining pregnancy-related causes of death from verbal autopsies. Popul Health Metr. 2007, 5: 1-10.1186/1478-7954-5-1. doi: 10.1186/1478-7954-5-1,View ArticlePubMedPubMed CentralGoogle Scholar
- Oti SO, Kyobutungi C: Verbal autopsy interpretation: a comparative analysis of the InterVA modelversus physician review in determining causes of death in the NairobiDSS. Popul Health Metr. 2010, 8: 21-10.1186/1478-7954-8-21. doi: 10.1186/1478-7954-8-21,View ArticlePubMedPubMed CentralGoogle Scholar
- Tadesse T, Demissie M, Berhane Y, Kebede Y, Markos A: Two-thirds of smear-positive tuberculosis cases in the community wereundiagnosed in northwest Ethiopia: population based cross-sectionalstudy. PLoS One. 2011, 6 (12): e28258-10.1371/journal.pone.0028258. doi: 10.1371/journal.pone.0028258,View ArticlePubMedPubMed CentralGoogle Scholar
- INDEPTH network: NDEPTH standardized verbal autopsy questionnaire. Available:http://www.Indepthnetwork.org/core_documents/indepthtools.htm.Accessed 2012 February 20,
- WHO: International classification of diseases (ICD). Available:http://www.who.int/classifications/help/icdfaq/en/index.html.Accessed 2012 February 22,
- Roberts C, Namee RM: Assessing the reliability of ordered categorical scales using kappa-typestatistics. Stat Methods Med Res. 2005, 14 (5): 493-514. 10.1191/0962280205sm413oa.View ArticlePubMedGoogle Scholar
- Tadesse S: Validating the InterVA model to estimate the burden of mortality from verbalautopsy data: a population-based cross-sectional study. PLoS One. 2013, 8 (9): e73463-10.1371/journal.pone.0073463. doi: 10.1371/journal.pone.0073463,View ArticlePubMedPubMed CentralGoogle Scholar
- Bauni E, Ndila E, Mochamah G, Nyutu G, Matata L, et al: Validating physician-certified verbal autopsy and probabilistic modeling(InterVA) approaches to verbal autopsy interpretation using hospital causesof adult deaths. Popul Health Metr. 2011, 9: 49-10.1186/1478-7954-9-49. doi: 10.1186/1478-7954-9-49,View ArticlePubMedPubMed CentralGoogle Scholar
- Yimer S, Bjune G, Yimaldu T, Hansen CH: Health-care seeking among pulmonary tuberculosis suspects and patients inrural Ethiopia: a community-based study. BMC Public Health. 2009, 9: 454-10.1186/1471-2458-9-454. doi: 10.1186/1471-2458-9-454,View ArticlePubMedPubMed CentralGoogle Scholar
- Wang J, Yang F, Xu B, Hongbing S: Gender difference in knowledge of tuberculosis and associated health-careseeking behaviors: a cross-sectional study in a rural area of china. BMC Public Health. 2008, 8: 354-10.1186/1471-2458-8-354. doi: 10.1186/1471-2458-8-354,View ArticlePubMedPubMed CentralGoogle Scholar
- Evans AC, Onifande D, Montoya R, Haro M, Alva J, et al: Gender-related factors influencing tuberculosis control in shantytowns: aqualitative study. BMC Public Health. 2010, 10: 381-10.1186/1471-2458-10-381. doi: 10.1186/1471-2458-10-381,View ArticlePubMedPubMed CentralGoogle Scholar
- Byass P, Ambruoso LD, Ouedraogo M, Qomariyah SN: Assessing the repeatability of verbal autopsy for determining cause of death:two case studies among women of reproductive age in Burkina Faso andIndonesia. Popul Health Metr. 2009, 7: 6-10.1186/1478-7954-7-6. doi: 10.1186/1478-7954-7-6,View ArticlePubMedPubMed CentralGoogle Scholar
- Chandramohan D: Verbal autopsy tools for adult deaths. [PhD thesis]. 2001, London: London school of hygiene and tropical medicineGoogle Scholar
- Mirza NM, Macharia WM, Wafula EM, Agwanda RO, Onyango FE: Verbal autopsy: a tool for determining cause of death in a community. East Afr Med J. 1990, 67: 693-8.PubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative CommonsAttribution License (http://creativecommons.org/licenses/by/2.0), whichpermits unrestricted use, distribution, and reproduction in any medium, provided theoriginal work is properly cited.