Since the introduction of combination antiretroviral therapy in 1996, there has been a substantial reduction in the incidence rates of death in the HIV-positive population.1–3 As the number of deaths from AIDS-related causes has dropped, causes of death have become more variable,4,5 with an increase in deaths from non-AIDS-related malignancies, liver, and cardiovascular disease, and other age-related comorbidities.6–10
It is important to monitor deaths in the HIV-infected population so that interventions can be targeted appropriately, should specific causes of death emerge or become predominant.11,12 Such monitoring is generally undertaken after classifying causes of death using the International Classification of Diseases system (ICD; WHO),13 the latest version of which (ICD-10) was endorsed in 1990. In contrast to ICD-9, the latest version includes more conditions that were considered to have HIV as an underlying cause, resulting in a far greater number of conditions being coded as “HIV-related”14 However, such an approach may not be optimal where the aim is to monitor emerging causes of death, particularly those that may be related to HIV or its treatment. In such a situation, it may be preferable to categorize causes of death according to organ system or etiology/pathology, rather than simply grouping all such death as being related to HIV.15–19
National surveillance of HIV-related mortality is generally performed using information from registries, with cause of death obtained from death certificates. However, it is well documented that death certificates alone, as well as site investigator self reporting, are often filled in erroneously (eg, with improper causal sequencing of diseases, or listing the mechanism of death rather than the underlying cause, which is insufficient to identify the underlying cause of death).20–22 In this situation, further adjudication is not possible because the relevant medical information on death circumstances has already been censored. Additionally, in an HIV-positive person,23,24 reporting of AIDS-related death from death certificates may be unreliable.25–27 Furthermore, given growing concerns about the long-term safety of antiretroviral treatment, accurate and reliable ascertainment of the underlying cause of death will have a direct impact on the study outcome.28 To facilitate the continuous surveillance of deaths in HIV-positive populations, and to permit an assessment of trends over time, there is an urgent need for a uniform classification system for deaths that occur in HIV-positive persons. Such a system must be accompanied by a structured and well-reported adjudication system for classifying causes of death.29–32
To address these needs, a meeting was held in Copenhagen in July 2004 to discuss the principles for determination of causes of death in HIV-infected persons. Participants were from scientific boards of several large cohort studies and ongoing randomized clinical trials that routinely collect data on causes of death. The Coding of Causes of Death in HIV (CoDe) project was launched, based on a joint agreement to project principles.33 We review the development and validation of the CoDe process, and present initial results from application of the process to deaths occurring in both the Data Collection on Adverse events of Anti-HIV Drugs Study and other external clinics.
The CoDe project includes the collection and ascertainment of detailed information on causes of death, and a central adjudication process. In the initial pilot phase, 12 HIV cohorts provided data on the CoDe case report form (CoDe form) for up to 10 most recent deaths that occurred from February 2003 until July 2004; 6 pairs of reviewers were assigned. Through this pilot phase, we tested a CoDe form, the CoDe Review case-report form (Review form), and CoDe guidelines.34 For the pilot study, the CoDe Working Group determined that 12 HIV cohorts would complete a CoDe form for the 5 most recent deaths, giving a total of 60 cases. Once the results from the pilot study were available, the results were evaluated by the CoDe Working Group, agreed upon changes were applied to the process, and a final version of the protocol and study documents was released in February 2005.33
Data Collection and Ascertainment
The 4-page CoDe form was used to collect information on background demographics, data source, risk factors, comorbidities, causes of death, autopsy results, combination antiretroviral therapy, laboratory measurements, and the potential role of serious toxicities in relation to death. Where available, autopsy reports were obtained. Specific instructions for the completion of the CoDe form have been provided,33 although the CoDe form is intended to be largely self explanatory. The integrated work processes of the CoDe project are outlined in the Figure. Office personnel at the Coordinating Center, including a specialist HIV clinician, assess the CoDe forms. Cases lacking relevant information or stating unknown death circumstances are queried, but all cases registered at the Coordinating Centre are processed to adjudication, regardless of this querying. During the querying process, centers receive feedback on the quality of the forms received, and further training is provided if necessary.
Central Adjudication Process
The final coding of cause of death is conducted through a Central Adjudication process performed independently by 2 randomly matched, external reviewers who remain anonymous to each other and who each follow specific guidelines.35 An online tool has been developed to facilitate this process. Currently, a total of 32 physician-reviewers from 22 European and one Australian clinic have been recruited.36 Based on information reported on the CoDe form, the external reviewers reconsider the sequence of events leading to death, and code one immediate cause, up to 4 contributing causes, and one underlying cause of death. Briefly, the immediate cause is the disease/injury that directly leads to death; contributing causes are those that contribute to the fatal outcome; and the underlying cause is the disease/injury that initiated the train of morbid events leading directly or indirectly to death.14 This definition is used by investigators reporting at the centers as well. A predefined categorization of causes (33 categories) is used.33 Categories 1–19 include specific causes of death known to be common in HIV-positive populations; categories 20–30 include more general categories and, if a specific or general category is not suitable, reviewers are able to use “other,” “unclassifiable,” or “unknown” codes. Reviewers are also asked to indicate their level of certainty to limit the “unknown” category and obtain more complete specific codes.
Both reviewers have to agree on the underlying cause of death (initial agreement cases). If there is an initial disagreement, an adjudication process is followed to reach consensus (Figure) (initial disagreement cases) and the reviewers are able to view their coreviewer's coding and correspond online via a comment box. If reviewers decide that they cannot achieve consensus, the case is referred to an additional reviewer for final adjudication.
The CoDe Project methodology described in this study was applied and tested in the Data Collection on Adverse events of Anti-HIV Drugs study during the period 2004–2008.37 This analysis includes the first 491 consecutive CoDe forms that went through the complete adjudication process before the end of February 2009. Each question on the form was judged as being completed if a response (eg, “Yes,” “No,” or “Unknown”) was present. Results of the review process were stratified according to initial agreement or disagreement. In statistical analysis, χ2 and Kruskal-Wallis tests were used for group comparisons. Univariable logistic regression models were used to identify factors associated with achieving initial agreement on the underlying cause of death by reviewers. The variables tested in these univariable models were age; body mass index (BMI); smoking; alcohol consumption; active injecting and noninjecting drug use; opiate substitution; ongoing dyslipidemia, hypertension and diabetes mellitus; prior cardiovascular disease; history of psychosis and depression; chronic hepatitis (hepatitis B virus, hepatitis C virus, hepatitis D virus); chronic elevation of liver transaminases; previous liver decompensation; liver failure 4 weeks before death; whether the death was sudden and unexpected; number of diagnoses that the patient had before death; status of antiretroviral treatment; most recent CD4 count and HIV RNA viral load before death; autopsy (yes or no); source of documentation; whether the reporting personnel had been involved in patient's care; and the final adjudicated cause of death. A multivariable model was developed by including all variables with P < 0.1 in univariable models. A stepwise selection method was used to confirm that all significant variables were included in the final model. A sensitivity analysis was carried out restricting the group to cases reported directly from the center. All analyses were performed using SAS version 9.1 (SAS Institute, Cary, NC).
A total of 491 reported deaths went through the review process and were assigned a final coding for the underlying cause of death. The median time from death to the receipt of the CoDe form at the coordinating center was 7.4 months (interquartile range [IQR] = 4.7–11.0). Deaths occurred from February 2000 to January 2007 (median, March 2005; IQR = July 2003–September 2005), with 135 (27%) deaths defined by investigators as “sudden.” The geographical distribution of deaths reflected that of the Data Collection on Adverse events of Anti-HIV Drugs study.37 Characteristics of the patients at the time of death are shown in Table 1. Autopsy was performed in 62 (13%) cases, and a summary was available for 51 (10%). In 17 (3%) cases, death was indicated by the reporting site as drug/treatment-related.
For 377 (77%) deaths, a standard CoDe form was completed by the participating center; for 114 (23%) information was not obtained directly from the centers but from the Data Collection on Adverse events of Anti-HIV Drugs database to accelerate the flow of the cases through the system, test online communication, and initiate review procedures. Of note, the latter had not routinely included information on use of methadone, alcohol, and noninjecting drug use, or history of psychiatric diseases; these cases were excluded from the evaluation of the completeness of CoDe forms.
Of the 377 CoDe forms received, 69 (30%) were completed by a person directly involved in the medical care of the patient. Almost two-thirds of these (n = 215) were completed by a physician, 19% (66) by a nurse, and 19% (64) by other personnel (eg, medical student). For 339 (90%) cases, information on source documentation was provided—forms had been completed using information from complete hospital files (67%), incomplete hospital files (19%), or outpatient clinic files (11%). Autopsy reports and information from other sources (obituary, relatives, general practitioner) were used in 3% of forms.
For the majority of questions considered, completeness was >90% (eTable 1, http://links.lww.com/EDE/A472).
Centers were asked in the form to provide the immediate and underlying condition that caused death, all HIV- and non-HIV-related diagnoses the patient had at the time of death, and a short narrative summary. The median number of diagnoses listed per death was 3 (IQR = 2–4). The underlying condition was completed for 135 (36%) cases. Among the 242 forms where the underlying condition was missing, 221 (91%) had conditions at the time of death listed, 230 (95%) had a brief narrative, and 210 (87%) had an immediate cause of death. There was no evidence that the difference between cases with and without information on the underlying condition was associated with position or involvement in the patient's care of the person completing the form. The underlying cause of death provided by the investigator and that received from the adjudication process agreed in 79 (58%) of 135 cases. In the remaining 56 cases, 16 cases had HIV as a cause of death as deemed by investigators but not as the adjudicated cause, while 4 deaths were listed as caused by HIV in the adjudicated cause but not by investigators.
Sections collecting information on treatment related to death and whether the death was sudden were often completed erroneously. Of the 22 cases where investigators had listed associations between the death and drugs/treatment, the coordinating center could confirm only 17 cases. Of the 135 deaths assigned by investigators as sudden, 61 (45%) were identified with a chronic, ongoing terminal condition that was subsequently assigned as the underlying cause of death by reviewers.
Central Adjudication Process Results
The review process assigned all 491 events a code for the underlying cause of death. A specific classification code was used for 424 events (86%) and a general classification code for 33 events (7%). Eight events (2%) were coded as other, reflecting the lack of an applicable code, and 26 (5%) as unknown, reflecting insufficient information to code the case (Table 2). The underlying cause of death was AIDS related for 32%, followed by hepatitis (17%), malignancies other than AIDS- or hepatitis-related (13%), non-AIDS infections (6%), substance abuse (5%), and myocardial infarction or other ischemic heart disease (4%).
There were 339 cases (69%) on which the 2 reviewers initially agreed and 152 (31%) where there was initial disagreement. In the initial-disagreement group, consensus was reached for 106 (22%) after reviewers viewed the coreviewer's coding, and the dialogue option was used in 46 cases (9%). In 136 (28%) cases, at least one reviewer proposed the underlying cause of death that was chosen as the final code, in the remaining 16 (3%) cases the final underlying cause of death was different from that initially proposed by either reviewer. For 35 (7%) initial disagreement cases, one reviewer had initially proposed other or unknown cause; in 19 (4%) of these cases an underlying cause of death was finally established (after viewing the coreviewer's evaluation), with a specific code allocated for 17 CoDe forms and a general code for the remaining two. None of the cases needed a third reviewer.
Reviewers classified the degree of certainty for the underlying cause of death in 479 cases, with 47% of cases classified as “definitely,” 30% as “likely,” and the remaining 22% as “possibly.” In 215 (63%) of initial-agreement cases, reviewers provided the same certainty (possibly, 23%; likely, 11%; definitely, 66%); in 93 (27%) cases they disagreed by one level of certainty; and in the remaining 31 cases (9%) they disagreed by 2 levels (ie, definitely and possibly).
The potential predictors of agreement on cause of death included in the final multivariable model were diabetes mellitus, hypertension, chronic HCV infection status, and history of depression (all as “Yes,” “No,” and “Unknown”), CD4 count and HIV RNA viral load (both most recent prior to death), age at the time of death, whether or not death was reported as sudden by investigator, number of diagnoses at the time of death (per increase of one diagnosis), cause of death agreed by reviewers, and the data source available for completion of CoDe form. Factors associated with agreement after adjustment were as follows: ongoing chronic hypertension, a history of depression, the number of diagnoses at the time of death, and the underlying cause of death from the review process (Table 3). For each additional diagnosis listed at the time of death, the odds of agreement were increased by approximately 20% (odds ratio [OR] = 1.19 [95% confidence interval (CI) = 1.05–1.35]). As compared with forms where deaths were ultimately deemed to be due to AIDS-related causes, the odds of agreement were more than 80% lower when deaths were ultimately deemed to be due to other non–AIDS-related (0.17 [0.08–0.37]) or undetermined causes (0.11 [0.04–0.36]). There was no difference in initial agreement of reviewers where deaths were ultimately deemed to be due to AIDS-related causes or as a result of chronic viral hepatitis, malignancy, cardiovascular disease, or violence. The odds of agreement between reviewers on cause of death were lower for patients with ongoing chronic hypertension (0.43 [0.22–0.85]) and a history of depression (0.43 [0.23–0.80]). In a sensitivity analysis, when the group was restricted to 377 cases reported directly from the center we found no substantial changes in the results mentioned earlier (eTable 2, http://links.lww.com/EDE/A472).
The reviewers needed an average of 15 minutes to review the CoDe form and complete the online Review form. The median time for finalizing the review was 4.3 months. If there was an initial disagreement between reviewers, an additional 0.7 month was required to achieve final consensus.
We describe a new tool for reporting and reviewing cause of death in HIV-positive subjects. The method is applicable to both clinical trials and observational studies, and is publicly available, free of charge. The initial results presented in this study document that the extent and format of the data collection are sufficient for an informed review, and that the coding scheme proposed with CoDe includes an adequate range of possible causes of death in HIV-positive patients.
CoDe has been already widely implemented by observational cohorts and clinical trials either in part or as a complete methodology. Currently, CoDe has been implemented in 2 observational studies (the Data Collection on Adverse events of Anti-HIV Drugs Study37 and the HIV/TB Project38) and by the International Network for Strategic Initiatives in Global HIV Trials (INSIGHT) group,39,40 for the collection and central adjudication of causes of death. In addition, the CoDe classification was used by The Antiretroviral Therapy Cohort Collaboration to describe data on causes of death provided by participating cohorts with different collection procedures.10
Although there was a high degree of completeness of the CoDe form, we identified sections where data were often omitted (underlying cause of death) or entered erroneously (sudden death, death related to medication). This illustrates a discrepancy between clinicians' and reviewers' perception of terms such as “sudden death” and “underlying cause of death,” supporting the necessity of an external adjudication process. It also shows the need to strengthen training systems through feedback to the clinics, Web-based learning tools, etc.
The importance of the adjudication process is underlined by the fact that in over two-thirds of cases, site investigators did not report an underlying condition that caused death, yet reviewers were able to do so. These events would not have been coded without central adjudication.
An additional advantage of the CoDe project is of the additional certainty regarding causes of death, which permits a wide range of analyses. For example, where it is necessary to be very specific about the cause of death, only definite causes could be included, whereas for sensitivity analyses it might be acceptable to relax the criteria to include deaths coded with greater uncertainty. Regardless of the application, the degree of certainty should always be included in reporting cause-of-death coding in scientific analyses.
During this pivotal stage of the project, we were able to identify a number of modifications that could help to further strengthen the data evaluation process and its efficiency. Some CoDe form sections, such as the autopsy report and medications related to death, could be simplified, shortening the form considerably without loss of accuracy. Central adjudication was the most time consuming and complicated part of the process. However, there was no need for adjudication between reviewers' decisions for the majority of events, suggesting that such cases could be assigned to only one reviewer. Additional analyses found that odds of agreement on the underlying cause of death were higher when the person had died of an AIDS event compared with deaths from other non-AIDS defining or undetermined causes. However, we were not able to identify a specific pattern that could predefine cases for which a single reviewer would suffice. The direct association between the number of premorbid diagnoses listed on the CoDe form and the chance of achieving agreement between reviewers suggests that completeness of recorded information and detailed description of the circumstances in which death occurred have the most impact in determining the underlying cause of death.
These observations suggest a better use of existing resources. By expanding the role of the central physician to include that of reviewer, and by allowing the review of the physician to be sufficient in “obvious” cases, the adjudication process may be initiated earlier in the project, making the process more efficient and decreasing the workload for external reviewers. These revised procedures will be evaluated prior to implementation and a 10% random sample will be sent out for external review as a quality control.
To conclude, CoDe is an initiative to improve the determination of the underlying cause of death. Strengths of this method include a centralized adjudication process, independent and clinically experienced reviewers, and the use of structured, comprehensive documents for the collection of prospective data based on predefined criteria. The diagnostic categorization incorporates information from a number of large recent HIV studies.4,6,8,10,39 The structure of the documents and the approach to coding underlying cause of death are familiar to reviewers as they build on conventional principles introduced internationally.14 With the increasing role of observational studies in the field of long-term safety of combination antiretroviral therapy, a standardized method should be used to ascertain the underlying cause of death.28,31,32 All CoDe documents are made publicly available free of charge. Having been tested and modified based on practical experience, the CoDe system provides a standardized, yet dynamic approach to analyze causes of death and evolving changes in mortality risk.
The CoDe working group is Antonella D'Arminio Monforte (ICONA) Italy, Åkerlinnd Börje (HivBIVUS) Sweden,* Genevieve Chene (Mortalité 2000; ART CC) France, Richard Davey (NIH) USA, Stephane De Wit (St Pierre Brussels Cohort) Belgium, Frank De Wolf (ATHENA) Netherlands, Matthias Egger (ART CC) United Kingdom, Michelle Ellefson (CHIP; EuroSIDA) Denmark, Wafaa El-Sadr (CPCRA) USA, Nina Friis-Møller (CHIP; DAD) Denmark, Christian Holkman Olsen (CHIP) Denmark,* Ole Kirk (CHIP; EuroSIDA) Denmark, Jesper Kjær (CHIP) Denmark, Justyna Kowalska (CHIP; EuroSIDA) Denmark, Matthew Law (AHOD) Australia, Bruno Ledergerber (SHCS) Switzerland, Charlotte Lewden (Mortalité 2000; ART CC) France, Jens Lundgren (CHIP; DAD) Denmark, Amanda Mocroft (Royal Free Hospital; EuroSIDA) United Kingdom, Maria Paulsen (CHIP) Denmark, Tim Peto (Oxford University; DART Study) United Kingdom, Andrew Phillips (Royal Free; DAD; EuroSIDA) United Kingdom, Christian Pradier (Nice Cohort) France, Mathias Bruyand (Aquitaine) France, Peter Reiss (ATHENA, AMC) Netherlands, Frank Rhame (ESPRIT) USA, Caroline Sabin (Royal Free; DAD) United Kingdom, Mateu Silvia (BASS) Spain,* Jonathan Sterne (ART CC) United Kingdom, Michael Saag (University of Alabama) USA, and Rainer Weber (SHCS) Switzerland. Former CoDe Working Group members indicated with.*
The CoDe reviewers are Elzbieta Bakowska (Wojewodzki Szpital Zakazny) Poland, Josip Begovac (University Hospital of Infectious Diseases) Croatia, Thomas Benfield (Hvidovre Hospital) Danmark, Paul Benn (Academic Department of Genitourinary Medicine) United Kingdom, Peter Bernhard (Copenhagen University Hospital) Denmark, Fabrice Bonnet (Hopital Saint-André) France, Mark Boyd (University of New South Wales) Australia, Antonella D′Arminio Monforte (Istituto Di Clinica Malattie Infettive e Tropicale) Italy, Andrea de Luca (Istituto di Clinica delle Malattie Infettive) Italy, Stéphane de Wit (Saint-Pierre Hospital) Belgium, Marc van der Valk (Academic Medical Center) The Netherlands, Ulrik Bak Dragsted (Copenhagen University Hospital) Denmark, Simon Edwards (Academic Department of Genitourinary Medicine) United Kingdom, Leo Flamholc (Malmö University Hospital) Sweden, Hansjakob Furrer (Inselspital Bern) Switzerland, Jose M. Gatell (Hospital Clinic i Provincial) Spain, Jannik Helweg-Larsen (Copenhagen University Hospital) Denmark, Djordje Jevtovic (The Institute for Infectious and Tropical diseases) Serbia, Ole Kirk (Copenhagen University Hospital) Denmark, Brygida Knysz (Wroclaw Medical University) Poland, Déborah Konopnicki (Saint-Pierre Hospital) Belgium, Ladislav Machala (Faculty Hospital Bulovka) Czech Republic, Nicolas Mueller (University Hospital, Zürich) Switzerland, Milos Opravil (University Hospital, Zürich) Switzerland, Marie-Christine Payen (Saint-Pierre Hospital) Belgium, Christian Pradier (Hôpital de l'Archet) France, Rainer Weber (University Hospital) Switzerland, Janos Szlavik (Szent Laszlo Hospital) Hungary, Vilma Uzdaviniene (Lithuania AIDS Center) Lithuania, Anna Vassilenko (Belarus State Medical University) Belarus, Saskia M.E. Vrouenraets (Academic Medical Center) The Netherlands, and Ian Williams (Academic Department of Genitourinary Medicine) United Kingdom.
The D:A:D study group includes Cohort PI's: W El Sadr* (CPCRA), G Calvo* (BASS), F Dabis* (Aquitaine), O Kirk* (EuroSida), M Law* (AHOD), A D'Arminio Monforte* (ICONA), L Morfeldt* (HivBIVUS), C Pradier* (Nice), P Reiss* (ATHENA), R Weber* (SHCS), S De Wit* (Brussels); Cohort coordinators and datamanagers: S Zaheri, L Gras (ATHENA), M Bruyand, S Geffard, (Aquitaine), K Petoumenos (AHOD), S Mateu, F Torres (BASS), M Delforge (Brussels), G Bartsch, G Thompsen (CPCRA), J Kjær, J Tverland (EuroSIDA), P Pezzotti (ICONA), E Fontas, C Caissotti (Nice), A Sundström, G Thulin (HivBIVUS), M Rickenbach (SHCS); Statisticians: CA Sabin*, AN Phillips*, Alim Kamara; Community representative: S Collins*; D:A:D coordinating office: SW Worm, N Friis-Møller, R Brandt, JD Lundgren*¢; Additional members: S Storfer*, G Pearce*, R Rode*. Steering Committee Members indicated with*; chair with ¢.
1.Mocroft A, Vella S, Benfield TL, et al. Changing patterns of mortality across Europe in patients infected with HIV-1. EuroSIDA Study Group. Lancet
2.van Sighem A, Danner S, Ghani AC, Gras L, Anderson RM, de WF. Mortality in patients with successful initial response to highly active antiretroviral therapy is still higher than in non-HIV-infected individuals. J Acquir Immune Defic Syndr
3.Keiser O, Taffe P, Zwahlen M, et al. All cause mortality in the Swiss HIV Cohort Study from 1990 to 2001 in comparison with the Swiss population. AIDS
4.Mocroft A, Brettle R, Kirk O, et al. Changes in the cause of death among HIV positive subjects across Europe: results from the EuroSIDA study. AIDS
5.Krentz HB, Kliewer G, Gill MJ. Changing mortality rates and causes of death for HIV-infected individuals living in Southern Alberta, Canada from 1984 to 2003. HIV Med
6.Lewden C, May T, Rosenthal E, et al. Changes in causes of death among adults infected by HIV between 2000 and 2005: the “Mortalite 2000 and 2005” Surveys (ANRS EN19 and Mortavic). J Acquir Immune Defic Syndr
7.Weber R, Sabin CA, Friis-Moller N, et al. Liver-related deaths in persons infected with the human immunodeficiency virus: the D:A:D study. Arch Intern Med
8.Monforte A, Abrams D, Pradier C, et al. HIV-induced immunodeficiency and mortality from AIDS-defining and non-AIDS-defining malignancies. AIDS
9.Palella FJ Jr, Baker RK, Moorman AC, et al. Mortality in the highly active antiretroviral therapy era: changing causes of death and disease in the HIV outpatient study. J Acquir Immune Defic Syndr
10.The Antiretroviral Therapy Cohort Collaboration. Causes of death in HIV-1-infected patients treated with antiretroviral therapy, 1996–2006: collaborative analysis of 13 HIV cohort studies. Clin Infect Dis.
11.Rosenthal E, Poiree M, Pradier C, et al. Mortality due to hepatitis C-related liver disease in HIV-infected patients in France (Mortavic 2001 study). AIDS
12.Hooshyar D, Hanson DL, Wolfe M, Selik RM, Buskin SE, McNaghten AD. Trends in perimortal conditions and mortality rates among HIV-infected patients. AIDS
13.Selik RM, Byers RH Jr, Dworkin MS. Trends in diseases reported on U.S. death certificates that mentioned HIV infection, 1987–1999. J Acquir Immune Defic Syndr
14.World Health Organization. International Statistical Classification of Diseases and Related Health Problems 10th Revision, Version for 2007. Available at: http://apps.who.int/classifications/apps/icd/icd10online/
. Accessed February 2009.
15.Strategies for Management of Anti-Retroviral Therapy/INSIGHT; DAD Study Groups. Use of nucleoside reverse transcriptase inhibitors and risk of myocardial infarction in HIV-infected patients. AIDS.
16.Sabin CA, Smith CJ, Youle M, et al. Deaths in the era of HAART: contribution of late presentation, treatment exposure, resistance and abnormal laboratory markers. AIDS
17.Mocroft A, Gatell J, Reiss P, et al. Causes of death in HIV infection: the key determinant to define the clinical response to anti-HIV therapy. AIDS
18.Reisler RB, Han C, Burman WJ, Tedaldi EM, Neaton JD. Grade 4 events are as important as AIDS events in the era of HAART. J Acquir Immune Defic Syndr
19.Smith CJ, Olsen CH, Mocroft A, et al. The role of antiretroviral therapy in the incidence of pancreatitis in HIV-positive individuals in the EuroSIDA study. AIDS
20.Johansson LA, Westerling R. Comparing hospital discharge records with death certificates: can the differences be explained? J Epidemiol Community Health
21.Roulson J, Benbow EW, Hasleton PS. Discrepancies between clinical and autopsy diagnosis and the value of post mortem histology: a meta-analysis and review. Histopathology
22.Villar J, Perez-Mendez L. Evaluating an educational intervention to improve the accuracy of death certification among trainees from various specialties. BMC Health Serv Res
23.Kowalska JD, Mocroft A, Ledergerber B, et al. A standardized algorithm for determining the underlying cause of death in HIV infection as AIDS or non-AIDS related: Results from the EuroSIDA study. HIV Clin Trials.
24.Petersen JL, Haque G, Hellkamp AS, et al. Comparing classifications of death in the Mode Selection Trial: agreement and disagreement among site investigators and a clinical events committee. Contemp Clin Trials
25.Barchielli A, Buiatti E, Galanti C, Giovannetti L, Acciai S, Lazzeri V. Completeness of AIDS reporting and quality of AIDS death certification in Tuscany (Italy): a linkage study between surveillance system of cases and death certificates. Eur J Epidemiol
26.Nojilana B, Groenewald P, Bradshaw D, Reagon G. Quality of cause of death certification at an academic hospital in Cape Town, South Africa. S Afr Med J
27.McCormick A. Unrecognised HIV related deaths. BMJ
28.Granger CB, Vogel V, Cummings SR, et al. Do we need to adjudicate major clinical events? Clin Trials
29.Coe AM, Tumilty SM, Graham C, et al. T1009 use of a death adjudication committee in a cohort study of hepatitis C and HIV infection. Gastroenterology
. 2008;134(4 suppl 1):A.
30.McGarvey LP, John M, Anderson JA, Zvarich M, Wise RA. Ascertainment of cause-specific mortality in COPD: operations of the TORCH Clinical Endpoint Committee. Thorax
31.Dechartres A, Boutron I, Roy C, Ravaud P. Inadequate planning and reporting of adjudication committees in clinical trials: recommendation proposal. J Clin Epidemiol
32.Naslund U, Grip L, Fischer-Hansen J, Gundersen T, Lehto S, Wallentin L. The impact of an end-point committee in a large multicentre, randomized, placebo-controlled clinical trial: results with and without the end-point committee's final decision on end-points. Eur Heart J
33.Coding Causes of Death in HIV Protocol Version 1.0. CoDe Website. Available at: http://www.cphiv.dk/Portals/_default/pdfhfolder/code_protocol_ver_1.0.pdf
. Accessed February 2005.
34.Olsen CH, Friis-Moller N, d'Arminio A, et al. Pilot of the CoDe (Coding of Death) project—a standardized approach to code cause of death in HIV infected individuals. In: 10th European AIDS Conference; 17 November 2005. Abstr PE18.4/9.
35.The CoDe Project review documents. The CoDe Project website. Available at: http://www.chip.dk/CoDe/Review/tabid/103/Default.aspx
. Accessed 20 April 2009.
36.The CoDe Project Reviewers list. The CoDe Project website. Available at: http://www.cphiv.dk/CoDe/CoDeReviewers/tabid/324/Default.aspx
. Accessed 27 November 2009.
37.Friis-Moller N, Weber R, Reiss P, et al. Cardiovascular disease risk factors in HIV patients—association with antiretroviral therapy. Results from the DAD study. AIDS
38.The HIV-TB Project. The HIV-TB Project website. Available at: http://www.cphiv.dk/HIVTB/tabid/284/Default.aspx
. Accessed February 2009.
39.Lifson AR, Belloso WH, Carey C, et al. Determination of the underlying cause of death in three multicenter international HIV clinical trials. HIV Clin Trials
40.International Network for Strategic Initiatives in Global HIV Trials website. START Study Protocol. Available at: http://insight.ccbr.umn.edu/official_documents/START/protocol_documents/START_Protocol.pdf
. Accessed 9 December 2008.