US Burden of Disease and Injury Study, 1996

AIDS and non-AIDS associated HIV infection

2. Selected invasive cancers

Diabetes mellitus and selected sequelae

4. Ischemic heart disease

Stroke
Non-fatal injuries
Congenital anomalies
Unipolar major depression and substance abuse

1.AIDS and Non-AIDS Associated HIV Infection

AIDS

Data Sources

Data on the incidence, prevalence and mortality associated with AIDS and AIDS related conditions were obtained from the HIV and AIDS surveillance systems maintained at the Centers for Disease Control and Prevention (CDC). Estimates are based on data reported to this data system through June 1999. The 1993 case definition was used to identify AIDS cases [1]. This definition classifies HIV-infected persons with CD4 T-lymphocyte counts <200/µL or CD4 percentage of lymphocytes <14% as AIDS cases even if they have not experienced an AIDS-defining opportunistic infection. The case counts were inflated by a factor of 10% to adjust for underreporting of AIDS cases [2]. In addition to adjusting for underreporting, the counts were adjusted for the reporting delays. This adjustment uses a maximum likelihood procedure that provides extra weight to recently reported cases and deaths as an estimate of the numbers that would eventually be reported. This procedure assumes that the reporting delay distribution does not vary over the 6 prior to most recent year for which reports are available [3].

Active surveillance of deaths among AIDS cases and death certificate data from the National Center for Health Statistics were used to generate the number of AIDS deaths. The counts from these data sources do not completely coincide because some AIDS cases die of other causes (e.g. injury), and some persons designated as dying of HIV may not have technically met the AIDS case definition prior to their death. Therefore, they would not have been reported to that surveillance system. However, if they truly died of HIV infection, it is highly likely that their CD4 T-lymphocyte count, if known, would have been low enough to meet the AIDS case definition. Therefore, for estimation of deaths due to AIDS, it was assumed that all death certificates with HIV infection listed as the underlying cause pertained to persons who died of AIDS. Also, death certificates with any mention of HIV infection were included if the certificate also identified an AIDS-defining illness, an illness likely to be due to underlying HIV infection (cardiomyopathy, aspergillosis, or nocardiasis) or cardiac arrest [4].

Estimation Procedures

The mortality, and to a lesser extent incidence, of AIDS began to decline rapidly in 1996. There is good evidence that the initial decline that began in 1995 and continued in 1996 was primarily the result of treatment of patients who had advanced immune suppression with multiple drug therapy that includes a Protease Inhibitor [5]. These regimens are known as highly active anti-retroviral therapies (HAART). Therefore, since the software used in the BODI methods assumes a steady state relationship between incidence, prevalence and case-fatality it is not possible to derive a plausible set of "DISMOD-consistent" estimates using available data on all these measures for AIDS from 1996. Hence the perspective adopted was to try and calculate an expected average duration of disease based on the information available in 1996 that was least influenced by the rapidly changing clinical situation that was emerging during 1996. Since HAART therapy at that time was primarily directed to very ill patients with AIDS, it was assumed that the death rates would decline faster than the incidence rate [6]. Therefore, age specific AIDS incidence and prevalence were entered in DISMOD to estimate average durations of AIDS while ignoring the resulting mortality figures. It was assumed that no patient experienced a remission from AIDS. In general the mortality rates in the resulting DISMOD models were higher than the observed values since case fatality rates were precipitously dropping during 1996. Therefore, the measured mortality rates have been entered in the epidemiologic estimation tables along with the "synthetic" average durations derived from the measured incidence and prevalence values. These probably represent reasonable estimates that would have been expected by most practitioners at that time when the impact of treatment was not completely appreciated. The "true" expected, average, duration will not be fully understood until more data are garnered from patients who have been under treatment since 1996.

Non-AIDS Related HIV-Infection

Data Sources and Estimation Procedures

Data on the prevalence of non-AIDS HIV-infections were obtained from CDC. They are based on a number of sentinel survey systems maintained by the CDC. The incidence values for AIDS were then entered in DISMOD as mortality figures since the onset of AIDS was the endpoint of interest in this model. Age specific average durations from infection until the onset of AIDS were obtained from two sources. The durations entered into DISMOD for persons ages 5-65 years were based on data published in the Collaborative Group on AIDS Incubation and HIV Survival Study [7]. Estimates of duration from infection to AIDS for children 0-5 years of age were derived from a published analysis of data from the Pediatric Spectrum of Disease Project [8]. Both of these analyses provided median time intervals for the durations of interest. Since multiple studies have suggested that the incubation periods for HIV infection follow a Weibull rather than a simple exponential distribution, the median values were used as the starting estimates of the average durations in these models [9]. However, since both of the studies that provided the incubation periods did not include the CD4 criteria in the AIDS case definitions, the assumption proffered by Holtgrave et al.was adopted and one year was subtracted from each of the age-specific duration values [10].

In order to obtain incidence numbers commensurate with the observed AIDS case rates and estimated prevalence numbers adjustments were made to the final durations (see table). None of these adjustments involved values greater than one year for any age group. The final number of total incident cases estimated from DISMOD was between 55,000 and 60,000 which approximates Rosenberg's estimate for the number of cases in the early 1990. The prevalence figures from this model closely approximated the numbers provided by CDC. All of these estimates seemed well within the 20% coefficient of variation observed in the published results of the back-calculation studies used to anchor these DISMOD models [11].

Original and final duration values from the literature used to model HIV and AIDS estimates.
Age Groups in Published Materials / Median time to Clinical AIDS (years) / Median time to AIDS adjusted for T-cell / DISMOD Age Groups / Time to AIDS Entered in DISMOD / Adjustments after DISMOD Modeling with rounding
0-4 / 5.0 / 4.0 / 0-4 / 4.0 / 4.0
5-14 / 11.0 / 10.0 / 5-14 / 10.0 / 10.0
15-24 / 11.0 / 10.0 / 15-24 / 10.0 / 10.0
25-34 / 9.8 / 8.8 / 25-44 / 8.2 / 9.0
35-44 / 8.6 / 7.6 / 45-64 / 5.5 / 6.0
45-54 / 7.7 / 6.7 / 65-74 / 4.8 / 5.0
55-64 / 5.3 / 4.3 / 75 / 3.4 / 4.0
65-74 / 4.0 / 3.0
75 / 3.0 / 2.0

2.Selected Invasive Cancers

Cancers Other Than Non-Melanoma Skin Cancer

Data Sources

The data sources for these estimates were the same as those used by most organizations that attempt to develop estimates for the descriptive epidemiology of cancer in the United States [12]. The estimated numbers of cases as well as five year survival proportions were calculated using cancer incidence rates from the regions of the United States included in the National Cancer Institute’s (NCI) Surveillance, Epidemiology, and End Results (SEER) program and population data collected by the US Bureau of the Census [13]. Mortality counts and rates were obtained from the Underlying Cause of Death data compiled by the National Center for Health Statistics [14].

Estimation Procedures

The software program SEER*Stat 2.0 that is provided with data from the NCI was used to generate the incidence and survival estimates used in this analysis [15].4 All estimates were obtained using the codes for invasive, malignant tumors from the 9 regions that have participated in the SEER program from 1973-1996. The anatomical coding system of the SEER*Stat 2.0 program was also used since it corresponds to the ICD-9 specifications used in the Global Burden of Disease (GBD) study [16]. Incidence rates were calculated using cancers diagnosed during the period 1992-1996.

The SEER data are sufficiently voluminous to provide cancer incidence rates for whites, blacks and the total population. However, there are insufficient numbers of persons of Asian race in these data to compute rates of satisfactory precision. Therefore, the ratio of the mortality rate in Asians to the mortality rate in the total population was multiplied by the incidence rate in the total population to provide an incidence estimate for Asians.

Survival curves computed using yearly intervals were generated using data from cancers diagnosed during the 10 year period of 1987-1996. Consistent with the methodology in the Global Burden of Disease Study it was assumed that patients surviving five years or longer after a cancer diagnosis were “cured”. Therefore, the average duration of disease entered into the DISMOD software was based on survival curves. In this analysis the average duration was estimated by applying standard life-table methods to annual, observed, survival proportions for the five years after diagnosis.

As noted above there are insufficient numbers of Asians in the SEER data to estimate survival rates or incidence rates. Since the procedure used to calculate incidence rates among Asians generally assumes that the relation between incidence and mortality is equivalent to that observed in the overall population, the duration of disease calculated for the all patients with cancer was used as the value for this variable in this population subgroup.

Mortality counts and rates were derived using the ICD-9 codes from the GBD. This coding system was applied to the underlying causes of death as recorded in the mortality records compiled by NCHS.

DISMOD

The above calculations provided estimates of incidence, mortality and average duration that were considered invariate when running the DISMOD software. Remission and case fatality rates were then varied within DISMOD until values for the three invariate estimates were equivalent to the values computed from the above data sources.

Non-Melanoma Skin Cancer

Data Sources and Estimation Procedures

Estimates for the 1996 incidence of basal cell and squamous cell carcinomas in the United States were derived using the methods described by Miller et al [17]. These incidence rates were summed and submitted as the estimates for non-melanoma skin cancers. Briefly, data from the special National Cancer Institute (NCI) survey of basal cell and squamous cell cancer in eight regions of the United States conducted between June 1, 1977 and May 31, 1978 were used to obtain a reasonably representative set of baseline incidence rates for the United States [18]. The incidence rates from these surveys were then extrapolated to1996 using the average annual rate of increase observed in two long-term continuous population-based non-melanoma skin cancer registries that had been in operation in North America since 1977. These registries were located in the province of British Columbia and the Kaiser-Permanente Health Maintenance Organization in Portland, Oregon. The increases observed in the British Columbia registry were larger than those observed in Portland. Therefore, separate extrapolations using each of these estimated rates of increase from 1978 to 1996were computed. The average of the two sets of rates for 1996 were used as the incidence estimates for persons of white race for the figures submitted to the Burden of Disease Unit.

Non-melanoma skin cancer is much less common in African-Americans than the rates observed in whites. Estimates for the incidence of this disease in Blacks in the United States were derived from the observation by Halder et al that these cancers are 68 times less frequent in African-Americans than in whites [19]. Therefore, we simply divided the incidence estimates for 1996 in whites by 68 to obtain estimates for Blacks. Incidence rates for Asians were assumed to be intermediate between those of whites and Blacks.

DISMOD

Since non-melanoma skin cancer is rarely fatal in the United States the assumption that a post diagnosis survival of five or more years represents cure was abandoned. Therefore, estimates of average duration and prevalence derived from DISMOD were computed by assuming that the incidence rates obtained from the above procedures, and mortality rates from the underlying cause of disease data, were invariate. It was also assumed that the remission rate for these cancers was at least 90%. With these assumptions as starting points, remission and mortality rates were varied until consistency was obtained between the incidence and mortality estimates. The resulting average durations were then assumed to represent reasonable estimates for this disease.

3.Diabetes Mellitus and Selected Sequelae

Data Sources

The data sources for these estimates were the National Health Interview Survey (NHIS), the National Health and Nutrition Examination Survey III (NHANES) and the NHANES I Epidemiologic Follow-up Survey (NHEFS). Estimates of self-reported diabetes prevalence were supplied by the Division of Diabetes Translation (DDT), CDC based on their analysis of data from the 1994-1996 NHIS as well as new analyses conducted for this study of the 1997 NHIS.

Provisional estimates of diabetes incidence were derived using answers to questions from the NHIS concerning whether the diabetes was diagnosed within 12 months of the date of the interview. Incidence rates for adolescents and children were derived using published estimates from population-based registries of insulin-dependent diabetes mellitus [20].

The self-reported estimates were compared to data from NHANES III to estimate the volume of undiagnosed diabetes. This survey contains questions similar to those in the NHIS on self-reported incidence and prevalence, as well as biochemical assays of fasting plasma glucose, and a 2 hour oral glucose tolerance test. To identify undiagnosed diabetics we used the American Diabetes Association criteria that only use fasting plasma glucose levels since the mortality associated with this approach is similar to that found with the more complex World Health Organization diagnostic standards [21].

Mortality data tapes in 1996 were used to identify the number of deaths attributed to ICD-9 code 250. The numbers and rates provided in the tables are based on these data. The number of deaths attributable (though not coded with diabetes as the underlying cause) to diabetes in DISMOD were calculated using published estimates on the age, sex and specific relative risk for total mortality associated with diabetes in adults [22]. These publications used NHEFS data. An independent analysis of this dataset with proportional hazard modeling was also conducted for this study. The predominant form of diabetes in children and adolescents is Type I, or insulin-dependent diabetes mellitus. Therefore relative-risks in children and adolescents (i.e < 24years of age) were based on published estimates from limited cohort studies which have demonstrated relative risk between 6-8, and the expert opinion of DDT staff [23]. These colleagues asserted that the relative risks increased with decreasing age. Incidence rates for sequelae were derived directly from the Global Burden of Disease (GBD) estimates.

Estimation Procedures

We estimated smoothed, age, race and sex-specific, self-reported prevalences for diabetes using quadratic logistic regression modeling of the NHIS and NHANES data among black and white racial groups. All the surveys indicated that the self-reported prevalence of diabetes decreased after the age of 65-74 years. However, experts in the DDT at the CDC indicated that these declines were probably the result of selection bias since household based surveys such as the NHIS do not include institutionalized persons. These experts felt that elderly persons with diabetes were more likely to be in nursing homes and other institutionalized settings. Therefore, they are not adequatelyrepresented in the NHIS and NHANES surveys. Therefore, we allowed the prevalence of diabetes to increase across all age groups, but held theincidence of disease in the age groups 65-74 and 75+ years relatively stable in the DISMOD models. Published results form the NHANES III data as well as analyses conducted by USBODI staff indicated that approximately 1/3 (33.8%) of diabetes is undiagnosed. Therefore, the self-reported prevalence of diabetes in adults was increased by a factor of 1.51 (i.e. self-reported prevalence/[1-0.338]). There were insufficient numbers of Asians and Native Americans in the national surveys to produce estimates for these groups. However, there were enough deaths attributed to diabetes in Asians to derive estimates for this group by assuming identical case-fatality relationships in this racial subgroup as there are in Whites, and adjusting the incidence rates to the registered death rates for this disease.