The rollout of antiretroviral therapy (ART) has decreased mortality in HIV-infected individuals in sub-Saharan Africa.1 Despite this, mortality remains high due to late HIV diagnosis, inadequate access to ART, and lack of diagnostic certainty with respect to opportunistic coinfections.2,3 Reliable information on the causes of death of HIV-infected individuals is needed to evaluate clinical management algorithms, to guide research, inform policy, and trace newly emerging diseases.4,5
Although a complete autopsy is the undisputed gold standard for establishing the cause of death, its practice is limited due to the need for well-trained staff and well-equipped health facilities, low acceptance rates, its mutilating nature, and the risk of disease transmission among others.6,7 Medical death certificates completed by the health care worker are often used as an alternative. However, the reliability when compared with complete autopsies is suboptimal and is expected to be worse in settings where access to diagnostics before death is limited.8,9
Therefore, other alternatives are studied including needle autopsy: postmortem percutaneous sampling of organs with a biopsy needle. This method overcomes some of the disadvantages of complete autopsy.10 In non–HIV-infected populations, needle autopsy accurately ascertained the cause of death in 43%–68% of cases when compared with complete autopsy.11,12 In HIV-infected patients, needle autopsies have been performed; however, a comparison with complete autopsy has never been published.13–15 In sub-Saharan Africa, needle autopsies have been mainly used in HIV-infected children to sample single organs and led to an organ-specific diagnosis in 62%–78%.16,17
A potential issue with needle autopsy is the lack of macroscopic evaluation. This could lead to sampling error, both of the target organ and any target lesion.11,14,15,18 Adding ultrasound guidance might overcome this. A South African study used ultrasound guidance to biopsy the spleen, kidneys, and heart, which led to successful biopsies in 87%, 100%, and 100%, respectively.13 However, HIV-infected individuals often die of disseminated infections and the likelihood of finding the cause of death in any of the sampled organs may be high.6,19
We sought to investigate needle autopsy, both blind and ultrasound guided, as a method to establish the cause of death in HIV-infected hospitalized patients in Uganda. We evaluated the agreement in cause of death between blind needle autopsy and complete autopsy and determined the added value of ultrasound guidance.
Setting and Population
The study was conducted from February until June 2013 in Mulago hospital, a tertiary level hospital in Kampala, Uganda. HIV-infected adults (aged >18 years) who died on one of the medicine wards were actively recruited by seeking written informed consent for study participation from their next of kin. In case an autopsy request reached the mortuary for a patient fulfilling the eligibility criteria from another hospital ward than the medicine wards, the next of kin was also approached for written informed consent. Patient inclusion took place from Monday to Saturday during working hours. For eligible patients dying outside this time frame, an attempt was made to reach the next of kin to ask consent. Postpartum deaths and deaths after trauma were excluded. The HIV serological status was based on documentation in the hospital chart. For feasibility reasons, a study period of 18 weeks with an anticipated number of 5 autopsies per week was used.
The needle autopsy took place within 4 hours after consent was granted in the hospital mortuary. One automated biopsy needle 14 G × 16 cm (Bard Max-Core disposable core biopsy instrument, Bard Inc., Covington, GA) was used per patient. The objective of the needle autopsy was to obtain adequate tissue for histological examination for each organ. The operator was free to perform as many punctures as needed; however, a minimum of 3 attempts and a reasonable maximum of approximately 10 attempts was set.
First, a blind needle autopsy was performed. Surface marking points and palpation were used to biopsy the brain, heart, liver, spleen, right and left kidneys, and right and left lungs. Then, an ultrasound-guided needle autopsy was performed using a pocketsize, portable ultrasound scan (Vscan V1.2; GE Healthcare, Horten, Norway) to biopsy the heart, liver, spleen, and both kidneys. The ultrasound scan was used for real-time guidance to the target organ. However, if ultrasound identified any lesion(s) within or outside the target organ, an attempt was made to also puncture that lesion. Finally, a complete autopsy was performed, including evisceration of the brain. After completing all study procedures, the body was embalmed. All tissue samples were fixed in 10% formalin, and hematoxylin and eosin (H&E) slides were made from each tissue section.
Two teams of pathologists reviewed the H&E-stained slides. Per individual patient, 1 team read the slides from the blind needle autopsy followed by the slides from the ultrasound-guided autopsy. The other team read the slides from the complete autopsy. Whether the team read the slides of the needle or the complete autopsy depended on who had performed the complete autopsy; if a member of team A had performed the complete autopsy, those slides were allocated to team A (Fig. 1). The pathologists were blinded for any information obtained during needle autopsy when judging the complete autopsy slides and vice versa. The pathologists could request additional stains, including Ziehl Neelsen, Grocott silver, periodic acid–Shiff, Giemsa, Gram, and immunohistochemistry as indicated for complete diagnosis. Criteria for tuberculosis (TB) diagnosis were the presence of acid-fast bacilli after Ziehl Neelsen staining or granuloma formation with the presence of giant cells or remarkable (caseous) necrosis not otherwise explained.
Each team agreed upon a maximum of 3 major diagnoses for the diseases directly contributing to death and a maximum of 5 minor diagnoses for the diseases not directly contributing to death. Ultrasound guidance was not used to biopsy the brain and the lungs. Therefore, the histological findings of the tissue obtained during ultrasound-guided autopsy were combined with the histological findings of the blind autopsy for brain and lungs. For example, if the major diagnosis was established in brain tissue, this diagnosis was also noted for the ultrasound-guided autopsy, even if the ultrasound-guided slides did not contain brain tissue. The pathologists were provided with the available clinical information. If a diagnosis was clinically established but not confirmed with autopsy, this diagnosis was not noted as outcome in any of the autopsy procedures.
Outcomes and Statistics
The primary outcome was concordance in major diagnosis between blind or ultrasound-guided needle autopsy (index test) and complete autopsy (gold standard). A major diagnosis was concordant if the exact disease was identified with the index test, for example, blind needle autopsy identifying pneumonia of unknown origin as major diagnosis and the complete autopsy identifying an Aspergillus pneumonia was considered as no concordance. We report both the proportion of major diagnoses that were identified with needle autopsy and the proportion of patients in which 1 or more major diagnoses were identified. We did subgroup analysis for the different disease categories: infections, malignancies, and other noncommunicable diseases.
When assessing the reason for not identifying the major diagnosis, 6 categories were defined (Box 1). When reporting the falsely identified major diagnoses, only diagnoses that were a different disease were considered. For example, if blind needle autopsy identified pneumonia of unknown origin as major diagnosis and the complete autopsy identified an Aspergillus pneumonia, we did not consider this a falsely identified major diagnosis.
- Sampling error organ: the needle autopsy did not obtain sufficient tissue for histological review of a target organ.
- Sampling error lesion: the needle biopsy did obtain sufficient tissue for histological review of a target organ but the tissue did not contain the major diagnosis involved.
- Major diagnosis in tissue not sampled in needle autopsy: the major diagnosis established with complete autopsy was identified in an organ that was not a target organ of the needle autopsy, for example, digestive tract in the case of gastrointestinal Kaposi's sarcoma.
- Difference in interpretation of histological findings: histological assessment of the tissue obtained with needle autopsy did show abnormality but the major diagnosis that was established with complete autopsy was not identified, for example, the blind needle autopsy identified pneumonia of unknown origin as major diagnosis and the complete autopsy identified Aspergillus pneumonia.
- Major diagnosis based on macroscopic evaluation: the major diagnosis established with complete autopsy was mainly based on the macroscopic evaluation, for example, a saddle pulmonary embolus.
- Minor diagnosis in needle autopsy: the major diagnosis established with complete autopsy was identified as a minor diagnosis in the needle autopsy.
Data were analyzed using STATA version 11.0 (Stata Corp, College Station, TX). Proportions are reported with 95% confidence intervals (CIs). To compare proportions, a χ2 test, McNemar's test, or a Fisher exact test was performed when appropriate. A P value <0.05 was considered statistically significant.
The study received ethical approval from the Joint Clinical Research Center Research and Ethics Committee (Uganda), the Mulago Internal Review Board (Uganda), and the Institute of Tropical Medicine Institutional Review Board (Belgium). The study received final approval and registration by the Uganda National Council of Science and Technology (HS 1300).
We obtained informed consent for 99 patients and performed 96 complete and blind needle autopsies and 95 ultrasound-guided needle autopsies. The median duration between death and autopsy procedures was 7 hours [interquartile range (IQR), 4–12]. Fifty-seven percent of patients were female, the median age was 35 years (IQR, 29–40), the median CD4 cell count was 47 cells per microliter (IQR, 17–165), 78% used cotrimoxazole prophylactic treatment, and 61% used ART. Complete autopsy revealed infections as a cause of death in 91% of patients, malignancies in 14%, and other noncommunicable diseases in 22% of patients (Table 1). TB was the main major diagnosis and was identified in 44% of patients.
Concordance Complete and Blind Needle Autopsy
Complete autopsy identified 121 major diagnoses in 92 patients; in 4 patients, no cause of death was identified. Concordance in the major diagnosis between blind needle and complete autopsy was 50% (95% CI: 40 to 59). No significant difference in concordance between the different disease categories was identified; concordance in infections was 54% (95% CI: 43 to 65), in malignancies 31% (95% CI: 9 to 61), and in other noncommunicable diseases 43% (95% CI: 22 to 66) (Table 2). TB, Cryptococcus neoformans infection, and Pneumocystis jirovecii pneumonia were the diagnoses with the highest concordance, in respectively, 71% (95% CI: 55 to 84), 69% (95% CI: 39 to 91), and 100% (95% CI: 29 to 100), with significantly higher concordance for TB when compared with all other diagnosis (P < 0.01).
Complete autopsy identified 2 major diagnoses in 27% of patients and 3 major diagnoses in 2%. In 46% (95% CI: 35 to 56) of patients, there was complete concordance for all major diagnoses and in 60% (95% CI: 49 to 70) for at least 1 major diagnosis. In patients with multiple major diagnoses, the second diagnosis was identified significantly less frequently than the first (19% versus 59%, P = 0.002).
The main reason for lack of concordance between the complete and blind needle autopsy was sampling error of the lesion (36%, 95% CI: 24 to 49) (Table 2). Thirty percent (95% CI: 19 to 43) of major diagnoses were missed because the major diagnosis was in an organ that was not sampled with the needle autopsy or because the macroscopic examination was needed to establish the major diagnosis.
Added Value of Ultrasound Guidance
Ultrasound-guided needle autopsy identified 4 additional major diagnoses; 3 true-positive diagnoses of TB and 1 false-positive diagnosis of hepatic failure. Therefore, concordance between ultrasound-guided needle autopsy and complete autopsy was 52% (95% CI: 42 to 61) for all major diagnoses, 57% (95% CI: 46 to 68) for all infectious major diagnosis, and 79% (95% CI: 63 to 90) for TB. In 46% (95% CI: 35 to 56) of patients, there was complete concordance for all major diagnoses. At least 1 major diagnosis was identified in 62% (95% CI: 51 to 72) of patients. None of these outcomes were significantly different compared with blind needle autopsy.
The 3 additional TB diagnoses were made in tissue of organs that were also successfully sampled in the blind needle autopsy, but did not contain the pathologic lesions. The false-positive diagnosis hepatic failure was identified as minor diagnosis in complete autopsy.
Blind needle autopsy identified 12 false-positive major diagnoses. Five of these (42%) were diseases that were considered a minor diagnosis in the complete autopsy: renal failure (n = 2), hepatic failure (n = 1), chronic aspecific meningitis (n = 1), and bacterial pneumonia (n = 1). In 3 patients (25%), blind needle autopsy identified 2 major diagnoses: bacterial pneumonia plus TB (n = 2) and plus Kaposi's sarcoma (n = 1). Complete autopsy confirmed the TB and Kaposi's sarcoma, but not the bacterial superinfections for these cases. In 2 patients (15%), blind needle autopsy missed the major diagnosis pulmonary TB, but falsely identified a bacterial pneumonia. Two diagnoses, intracerebral hemorrhage and bacterial pneumonia were identified with blind needle autopsy while complete autopsy did not identify any pathological lesion in the involved organs for these patients. For none of the bacterial pneumonias, Gram staining was requested.
Yield per Organ
Major diagnoses identified with complete autopsy were mainly present in the lung (63%, 95% CI: 54 to 72), the liver (44%, 95% CI: 35 to 53), and the spleen (34%, 95% CI: 25 to 43), excluding the 7 major diagnoses that needed macroscopic examination for confirmation (Table 3). This was the same for the different disease categories, although other noncommunicable diseases showed a trend to appear more often in the liver compared with the other categories (65%, 95% CI: 39 to 90, P = 0.06). TB was mainly present in the lungs (83%, 95% CI: 72 to 95), the liver (71%, 95% CI: 57 to 86), and the spleen (69%, 95% CI: 54 to 84).
Blind needle autopsy was most successful in identifying those major diagnoses that were evident in the liver (76%, 95% CI: 62 to 87) and the spleen (82%, 95% CI: 66 to 92). Only 40% (95% CI: 29 to 53) of the major diagnoses located in the lungs (Fig. 2) were identified. TB located in the lungs was identified with blind needle autopsy in 40% (95% CI: 24 to 58), in the liver in 73% (95% CI: 54 to 88), and in the spleen in 79% (95% CI: 60 to 92).
This is the first study to evaluate the concordance in cause of death between (ultrasound guided) needle autopsy and complete autopsy in HIV-infected adults. Blind needle autopsy identified half of the major diagnoses. For TB, the most prevalent cause of death, the concordance was significantly higher. The addition of ultrasound guidance did not improve the performance of needle autopsy. Most major diagnoses were located in the lungs; however, needle biopsy most successfully identified major diagnoses in the liver and spleen.
The concordance we found is comparable with the results obtained with blind needle autopsies in HIV-uninfected adults; however, 1 study found concordance in 67% of all major diagnoses.11,20,21 In this study, needle biopsies of heart, lungs, liver, and kidneys were combined with cultures of spleen and lungs for all patients and bone marrow biopsies on indication. Without these additional tests, concordance was 51%.20
The main reason for discordance was the absence of disease in the tissue obtained with needle biopsy. This can be the result of too few samples per organ, too little variation in sampling location within the organs and/or the lack of visual evaluation to guide the sampling. The addition of ultrasound guidance was intended to overcome this last issue. Portable ultrasound machines are increasingly available at relatively low costs and are therefore the imaging mode most easily applied in a resource-limited setting.22 However, we did not find a significant increase in concordance with ultrasound guidance. Other imaging modalities including computed tomography or magnetic resonance imaging have been studied, mainly as a noninvasive replacement of complete autopsies.23,24 For now, these modalities seem a long way from being feasible in resource-limited settings.
Another way to increase the performance of the needle autopsy could be the addition of microbiology. A South African study that combined (ultrasound guided) needle autopsy with microbiology identified 7 of 101 diagnoses exclusively with cultures, and a definite etiologic pathogen was found in another 5 patients.13 A study from Zimbabwe sampled lung tissue of 24 children and found 9 positive long tissue cultures (6 times gram-negative and 3 times gram-positive bacteria). The blood culture identified the same bacteria in 4 cases. Agreement with the histology and the clinical findings were not reported.17 Even in complete autopsy series, microbiology revealed diseases that remained histologically and clinically undiagnosed like Salmonella sp. sepsis.25 However, obtaining cultures postmortem also has disadvantages; adding extra tests offsets the simplicity and noninvasive nature of the needle autopsy, overgrowth of endogenous bacterial flora may render the microbiology results difficult to interpret and their yield might be limited in a setting where many patients receive broad-spectrum antibiotics in the 24 hours before death.
An important proportion (30%) of major diagnoses were missed because of reasons intrinsic to the procedure as we performed it; macroscopic findings were needed or the major diagnosis was established in tissue not sampled in the needle biopsy procedure such as the meninges, gastrointestinal tract, and lymph nodes. These organs are not easily obtained with blind needle autopsy. To improve the method, needle biopsy protocols that include clinical symptoms could be formulated to guide the sampling. For example, in a patient with symptoms of meningitis, an effort should be made to obtain meningeal tissue.
The majority of false-positive diagnoses were bacterial pneumonias. In bedridden, hospitalized patients defense mechanisms of the upper airway are less functional. Therefore, some degree of pulmonary inflammation is probably present in a large proportion of patients. The presence of a small mixed inflammatory infiltrate with neutrophilic granulocytes will prompt a pathologist immediately toward a diagnosis of pneumonia. However, because the extend of the pneumonia is difficult to estimate in small tissue cores as obtained with needle biopsy this may lead to false-positive results.
Although the groups were small and the differences were nonsignificant, there seems to be a trend that macroscopic judgment is most important in the diagnosis of other noncommunicable diseases. In a global transition period where noncommunicable diseases become more important, both in non–HIV-infected and (treated) HIV-infected patients, this could be of importance.4,26
Concordance in the second major diagnosis was significantly lower than in the first. We know that hospitalized HIV-infected patients often have multiple diseases when they die.19,25,27 Therefore, a high suspicion for more than 1 major diagnosis is needed in these patients, also in histopathological review.
Our study has several shortcomings. To prevent carryover effect, per individual patient, the slides of needle and complete autopsy were read by different teams. Although both teams consisted of experienced pathologists, differences in pathological judgment rather than differences in the tissue obtained cannot be ruled out. Moreover, the differentiation between major and minor diagnosis is subjective and further complicated by limited availability of clinical information. The large proportion of discordances in major and minor diagnosis reflects this. Finally, we used relatively stringent criteria to decide if concordance was present, and we took major diagnosis like unspecified pulmonary congestion into account. Therefore, our estimates of concordance are conservative.
We have determined the diagnostic value of needle autopsy in HIV-infected adults in Uganda. With a concordance rate of approximately 50%, one may question if this method is of value. This will depend on the setting and the aim of the needle biopsy. In settings where complete autopsy is not available, currently, cause of death is based on death certification and/or verbal autopsy. The accuracy of death certification is poor when compared with complete autopsy in HIV-infected patients.6,28,29 Verbal autopsies have so far only been validated against medical certificates.30,31 Needle autopsy increased accuracy and improves diagnostic certainty by providing histologic evidence for a certain disease. For example, in the case of TB, we found that 50% (n = 21) of the complete autopsy confirmed TB cases were not on TB treatment premortem. Needle autopsy identified 76% (n = 16) of these untreated TB cases. This indicates the high yield of needle autopsy for a frequent disease like TB when compared with clinical diagnosis.
In conclusion, blind needle autopsies are more accurate than clinical diagnosis but do not approach the accuracy of complete autopsies. Targeted needle biopsy protocols based on clinical symptoms or the addition of microbiology might improve the diagnostic yield. However, when adding extra steps, one must keep in mind that needle autopsies should remain easy to perform, acceptable, and safe to be a valid alternative for complete autopsy.
The authors thank all the relatives who provided consent for study participation. Moreover, they thank Moses Kamya, Head of the Department of Medicine, Makerere University, and the staff of the medicine wards of Mulago Hospital. They also thank Dr. Male-Mutumba, Dr. Asafu Munema, and Dr. Eva Mbwilo, the staff of the Mulago Hospital mortuary, in particular Mr. Edrisa Katende and the staff of the Makerere University Pathology laboratory, in particular Miss Betty Namwase. They thank prof. Lammens, head of the Pathology Department of the University of Antwerp and Christophe Hermans.
1. UNAIDS. UN Joint Programme on HIV
/AIDS (UNAIDS), Global Report: UNAIDS Report on the Global AIDS Epidemic 2013. November 2013, ISBN 978-92-9253-032-7. Available at: http://unaids.org
. Accessed June 26, 2014.
2. Biadgilign S, Reda AA, Digaffe T. Predictors of mortality
infected patients taking antiretroviral treatment in Ethiopia: a retrospective cohort study. AIDS Res Ther. 2012;9:15.
3. Lawn SD, Harries AD, Anglaret X, et al.. Early mortality
among adults accessing antiretroviral treatment programmes in sub-Saharan Africa. AIDS. 2008;22:1897–1908.
4. Lozano R, Naghavi M, Foreman K, et al.. Global and regional mortality
from 235 causes of death for 20 age groups in 1990 and 2010: a systematic analysis for the Global Burden of Disease Study 2010. Lancet. 2012;380:2095–2128.
5. Vogel G. Global health. How do you count the dead? Science. 2012;336:1372–1374.
6. Cox JA, Lukande RL, Lucas S, et al.. Autopsy causes of death in HIV
-positive individuals in sub-Saharan Africa and correlation with clinical diagnoses. AIDS Rev. 2010;12:183–194.
7. Burton JL, Underwood J. Clinical, educational, and epidemiological value of autopsy. Lancet. 2007;369:1471–1480.
8. Shojania KG, Burton EC, McDonald KM, et al.. Changes in rates of autopsy-detected diagnostic errors over time: a systematic review. JAMA. 2003;289:2849–2856.
9. Thurnheer R, Hoess C, Doenecke C, et al.. Diagnostic performance in a primary referral hospital assessed by autopsy: evolution over a ten-year period. Eur J Intern Med. 2009;20:784–787.
10. Turner GD, Bunthi C, Wonodi CB, et al.. The role of postmortem
studies in pneumonia etiology research. Clin Infect Dis. 2012;54(suppl 2):S165–S171.
11. Foroudi F, Cheung K, Duflou J. A comparison of the needle biopsy
post mortem with the conventional autopsy. Pathology. 1995;27:79–82.
12. Garg S, Punia RP, Basu S, et al.. Comparison of needle autopsy
with conventional autopsy in neonates. Fetal Pediatr Pathol. 2009;28:139–150.
13. Wong EB, Omar T, Setlhako GJ, et al.. Causes of death on antiretroviral therapy: a post-mortem study from South Africa. PLoS One. 2012;7:e47542.
14. Guerra I, Ortiz E, Portu J, et al.. Value of limited necropsy in HIV
-positive patients. Pathol Res Pract. 2001;197:165–168.
15. Baumgart KW, Cook M, Quin J, et al.. The limited (needle biopsy
) autopsy and the acquired immunodeficiency syndrome. Pathology. 1994;26:141–143.
16. Jeena PM, Coovadia HM, Chrystal V. Pneumocystis carinii and cytomegalovirus infections in severely ill, HIV
-infected African infants. Ann Trop Paediatr. 1996;16:361–368.
17. Nathoo KJ, Gondo M, Gwanzura L, et al.. Fatal Pneumocystis carinii pneumonia in HIV
-seropositive infants in Harare, Zimbabwe. Trans R Soc Trop Med Hyg. 2001;95:37–39.
18. Wellmann K. The needle autopsy
. A retrospective evaluation of 394 consecutive cases. Am J Clin Pathol. 1969;52:441–444.
19. Cox JA, Lukande RL, Nelson AM, et al.. An autopsy study describing causes of death and comparing clinico-pathological findings among hospitalized patients in Kampala, Uganda. PLoS One. 2012;7:e33685.
20. Huston BM, Malouf NN, Azar HA. Percutaneous needle autopsy
sampling. Mod Pathol. 1996;9:1101–1107.
21. West M, Chomet B. An evaluation of needle necropsies. Am J Med Sci. 1957;234:554–560 passim.
22. Spencer JK, Adler RS. Utility of portable ultrasound in a community in Ghana. J Ultrasound Med. 2008;27:1735–1743.
23. Roberts IS, Benamore RE, Benbow EW, et al.. Post-mortem imaging as an alternative to autopsy in the diagnosis of adult deaths: a validation study. Lancet. 2012;379:136–142.
24. Roberts IS, Traill ZC. Minimally invasive autopsy employing post-mortem CT and targeted coronary angiography: evaluation of its application to a routine Coronial service. Histopathology. 2014;64:211–217.
25. Martinson NA, Karstaedt A, Venter WD, et al.. Causes of death in hospitalized adults with a premortem diagnosis of tuberculosis
: an autopsy study. AIDS. 2007;21:2043–2050.
26. Julius H, Basu D, Ricci E, et al.. The burden of metabolic diseases amongst HIV
positive patients on HAART attending the Johannesburg Hospital. Curr HIV
27. Rana FS, Hawken MP, Mwachari C, et al.. Autopsy study of HIV
-1-positive and HIV
-1-negative adult medical patients in Nairobi, Kenya. J Acquir Immune Defic Syndr. 2000;24:23–29.
28. Ordi J, Ismail MR, Carrilho C, et al.. Clinico-pathological discrepancies in the diagnosis of causes of maternal death in sub-Saharan Africa: retrospective analysis. PLoS Med. 2009;6:e1000036.
29. Nelson AM, Perriens JH, Kapita B, et al.. A clinical and pathological comparison of the WHO and CDC case definitions for AIDS in Kinshasa, Zaire: is passive surveillance valid? AIDS. 1993;7:1241–1245.
30. Fligner CL, Murray J, Roberts DJ. Synergism of verbal autopsy and diagnostic pathology autopsy for improved accuracy of mortality
data. Popul Health Metr. 2011;9:25.
31. Lozano R, Lopez AD, Atkinson C, et al.. Performance of physician-certified verbal autopsies: multisite validation study using clinical diagnostic gold standards. Popul Health Metr. 2011;9:32.