Cause-specific Models Research Articles

Risk of childhood mortality in family members of men with poor semen quality.

What is the familial childhood mortality in first-degree (FDR) and second-degree relatives (SDR) of patients undergoing semen analysis (SA)? The relationship between infertility and congenital malformations (CM) in offspring is complex, with an increased risk of death due to CM in FDR, but not SDR, of men with lower semen parameters. Semen quality is an established predictor of men's somatic health. We can gain a better understanding of possible genetic or environmental determinants of the infertility phenotype by exploring familial aggregation of childhood mortality in relatives of men with poor semen quality. Retrospective cohort study from the Subfertility, Health and Assisted Reproduction study (cohort compiled 1996-2011) linked with patient/familial information from the Utah Population Database (UPDB). Index cases included a clinic-referred sample of 12 889 men who underwent SA and had adequate familial and follow-up data in the UPDB. Parameters of semen quality included: semen concentration, sperm count, motility, total motile count, sperm head morphology, sperm tail morphology and vitality. SA data were collected from two tertiary medical center andrology laboratories that have captured ~90% of all SA performed in Utah since 2004. Age- and sex-matched fertile controls were selected to create the comparison group for determining risk of childhood death (to age 20 years) in family members. A total of 79 750 siblings and 160 016 aunts/uncles were used to investigate the familial aggregation of childhood mortality. The main outcome was childhood mortality in FDR and SDR of men with SA and their matched controls. All-cause and cause-specific Cox proportional hazard models were used to test the association between semen quality and childhood mortality in family members. Cause-specific models were considered for cancer and CM. In the cohort of men with SA, there were 406 (1.0%) deaths in FDR and 772 (1.1%) deaths in SDR due to any cause. There was no significant difference in the risk of all-cause childhood mortality between the relatives of men with SA and the fertile control group [hazard ratio (HR)Female = 1.08, 95% CI = 0.88, 1.32; HRMale = 0.88, 95% CI = 0.75, 1.04]. We found no association between semen quality and risk for childhood cancer mortality in FDR or SDR (HRFDR = 0.98, 95% CI = 0.62, 1.54; HRSDR = 1.12, 95% CI = 0.83, 1.50). The FDR of men with SA and fertile controls were followed on average for 19.71 and 19.73 years, respectively. During this period of follow-up, FDR of men with SA had an unadjusted 40% relative risk of increased CM-related death. After stratifying by semen parameters and adjusting for birth year, we found FDR of men with worse semen quality, and notably azoospermic men (HR = 2.69, 95% CI = 1.24,5.84), were at higher risk of CM-related death. A large proportion of men with SA in the study had normal semen parameters. It is important to note that these men themselves may not be subfertile, but they were subfertile at the couple level (i.e. the female partner may be infertile). In addition, care is needed when interpreting our results, as we do not have semen measures on our sample of fertile men. Second, we were unable to include potential confounders such as medical comorbidities, smoking status, or environmental exposures. Third, men with SA were seen at the University of Utah or Intermountain Health Care clinics for a fertility evaluation thereby suggesting a more select population. Fourth, we chose to categorize morphology into equally distributed quartiles as a response to the fact that the World Health Organization threshold for normal motility changed multiple times during our study period. Lastly, we do not know the proportion of female partners with diagnosed infertility. We chose not to subcategorize each infertile male by infertile diagnosis because our goal was to understand how semen parameters influenced familial childhood mortality. We are not the first study to show a relationship between fertility and CMs. Children conceived through ART may be at higher risk of birth defects, however it is not known if the relationship is causal or if there is some underlying factor linking infertility and birth outcomes. This study provides further evidence that the increased risk of congenital birth defects may not be due to the ART, but rather genetic or environmental factors that link the two outcomes. We encourage further research in order to confirm a relationship between semen quality and increased risk for CM. This work was supported by the National Institutes of Health - National Institute of Aging [Grant numbers 1R21AG036938-01, 2R01 AG022095 and 1K12HD085852-01]. Authors have no competing interests to disclose. Not applicable.

Association of serum bilirubin levels with mortality in patients with diabetes initiating chronic hemodialysis: a competing risks analysis of a single-center cohort

Evidence remains scarce regarding the property of serum bilirubin levels as a prognostic factor for the incidence of mortality in dialysis patients. Here, we aimed to examine the association between serum bilirubin levels and mortality in patients with diabetes initiating hemodialysis. This was a single-center, observational, historical cohort study of 395 Japanese patients with diabetes initiating chronic hemodialysis. There were 98 women and 297 men, and the mean (±SD) age was 61 ± 12 years. Patients were dichotomized at the baseline median level of serum bilirubin (0.3 mg/dL). The endpoints were the incidence of all-cause and cause-specific mortality. Hazard ratios for these endpoints were estimated using the Fine-Gray subdistribution hazard model and cause-specific hazard model because of the presence of competing risks. During a median follow-up period of 3.9 years (range; 0.0–10.7 years), there were 83 cases of death, consisting of 39 and 44 cases from atherosclerotic cardiovascular diseases (ASCVDs) and non-ASCVDs, respectively. In the multivariate competing risks analysis, patients with lower bilirubin levels were associated with higher risk of ASCVD-related mortality (subdistribution hazard ratio 3.57, p = 0.015) but not with non-ASCVD-related mortality (0.83, p = 0.665) than those with higher bilirubin levels. Analysis treating bilirubin levels as a continuous variable, as well as cause-specific models yielded the similar results. Lower serum levels of bilirubin may be predictive of higher risk of ASCVD-related mortality in patients with diabetes initiating chronic hemodialysis. There was no relationship between levels of bilirubin and non-ASCVD-related mortality.

Impact of Socioeconomic Status on Patients Supported With a Left Ventricular Assist Device: An Analysis of the UNOS Database (United Network for Organ Sharing).

Low socioeconomic status (SES) is a known risk factor for heart failure, mortality among those with heart failure, and poor post heart transplant (HT) outcomes. This study sought to determine whether SES is associated with decreased waitlist survival while on left ventricular assist device (LVADs) support and after HT. A total of 3361 adult patients bridged to primary HT with an LVAD between May 2004 and April 2014 were identified in the UNOS database (United Network for Organ Sharing). SES was measured using the Agency for Healthcare Research and Quality SES index using data from the 2014 American Community Survey. In the study cohort, SES did not have an association with the combined end point of death or delisting on LVAD support (P=0.30). In a cause-specific unadjusted model, those in the top (hazard ratio, 1.55; 95% confidence interval, 1.14-2.11; P=0.005) and second greatest SES quartile (hazard ratio 1.50; 95% confidence interval, 1.10-2.04; P=0.01) had an increased risk of death on device support compared with the lowest SES quartile. Adjusting for clinical risk factors mitigated the increased risk. There was no association between SES and complications. Post-HT survival, both crude and adjusted, was decreased for patients in the lowest quartile of SES index compared with all other SES quartiles. Freedom from waitlist death or delisting was not affected by SES. Patients with a higher SES had an increased unadjusted risk of waitlist mortality during LVAD support, which was mitigated by adjusting for increased comorbid conditions. Low SES was associated with worse post-HT outcomes. Further study is needed to confirm and understand a differential effect of SES on post-transplant outcomes that was not seen during LVAD support before HT.

Investigating Rates of Hunting and Survival in Declining European Lapwing Populations.

Understanding effects of harvest on population dynamics is of major interest, especially for declining species. European lapwing Vanellus vanellus populations increased from the 1960s until the 1980s and declined strongly thereafter. About 400,000 lapwings are harvested annually and it is thus of high conservation relevance to assess whether hunting was a main cause for the observed changes in lapwing population trends. We developed a multi-event cause-specific mortality model which we applied to a long-term ring-recovery data set (1960–2010) of > 360,000 records to estimate survival and cause-specific mortalities. We found no temporal change in survival over the last 50 years for first-year (FY) and older birds (after first-year; AFY) originating from different ringing areas. Mean survival was high, around 0.60 and 0.80 for FY and AFY individuals, respectively. The proportion of total mortality due to hunting was <0.10 over the study period and the estimated proportion of harvested individuals (kill rate) was <0.05 in each year. Our result of constant survival indicates that demographic processes other than survival were responsible for the pronounced change in lapwing population trends in the 1980s. Our findings lend support to the hypothesis that hunting was not a significant contributor to the large-scale decline of lapwing populations. To halt the ongoing decline of European lapwing populations management should focus on life history stages other than survival (e.g. productivity). Further analyses are required to investigate the contribution of other demographic rates to the decline of lapwings and to identify the most efficient conservation actions.

Effect of alcohol consumption on all-cause and liver-related mortality among HIV-infected individuals.

The aim of the study was to examine the association between levels of past and current alcohol consumption and all-cause and liver-related mortality among people living with HIV (PLWH). A prospective cohort study of 1855 PLWH in Baltimore, MD was carried out from 2000 to 2013. We ascertained alcohol use by (1) self-report (SR) through a computer-assisted self interview, and (2) medical record abstraction of provider-documented (PD) alcohol use. SR alcohol consumption was categorized as heavy (men: > 4 drinks/day or > 14 drinks/week; women: > 3 drinks/day or > 7 drinks/week), moderate (any alcohol consumption less than heavy), and none. We calculated the cumulative incidence of liver-related mortality and fitted adjusted cause-specific regression models to account for competing risks. All-cause and liver-related mortality rates (MRs) were 43.0 and 7.2 per 1000 person-years (PY), respectively. All-cause mortality was highest among SR nondrinkers with PD recent (< 6 months) heavy drinking (MR = 85.4 deaths/1000 PY) and lowest among SR moderate drinkers with no PD history of heavy drinking (MR = 23.0 deaths/1000 PY). Compared with SR moderate drinkers with no PD history of heavy drinking, SR nondrinkers and moderate drinkers with PD recent heavy drinking had higher liver-related mortality [hazard ratio (HR) = 7.28 and 3.52, respectively]. However, SR nondrinkers and moderate drinkers with a PD drinking history of > 6 months ago showed similar rates of liver-related mortality (HR = 1.06 and 2.00, respectively). Any heavy alcohol consumption was associated with all-cause mortality among HIV-infected individuals, while only recent heavy consumption was associated with liver-related mortality. Because mortality risk among nondrinkers varies substantially by drinking history, current consumption alone is insufficient to assess risk.

Predicting Survival from Telomere Length versus Conventional Predictors: A Multinational Population-Based Cohort Study.

Telomere length has generated substantial interest as a potential predictor of aging-related diseases and mortality. Some studies have reported significant associations, but few have tested its ability to discriminate between decedents and survivors compared with a broad range of well-established predictors that include both biomarkers and commonly collected self-reported data. Our aim here was to quantify the prognostic value of leukocyte telomere length relative to age, sex, and 19 other variables for predicting five-year mortality among older persons in three countries. We used data from nationally representative surveys in Costa Rica (N = 923, aged 61+), Taiwan (N = 976, aged 54+), and the U.S. (N = 2672, aged 60+). Our study used a prospective cohort design with all-cause mortality during five years post-exam as the outcome. We fit Cox hazards models separately by country, and assessed the discriminatory ability of each predictor. Age was, by far, the single best predictor of all-cause mortality, whereas leukocyte telomere length was only somewhat better than random chance in terms of discriminating between decedents and survivors. After adjustment for age and sex, telomere length ranked between 15th and 17th (out of 20), and its incremental contribution was small; nine self-reported variables (e.g., mobility, global self-assessed health status, limitations with activities of daily living, smoking status), a cognitive assessment, and three biological markers (C-reactive protein, serum creatinine, and glycosylated hemoglobin) were more powerful predictors of mortality in all three countries. Results were similar for cause-specific models (i.e., mortality from cardiovascular disease, cancer, and all other causes combined). Leukocyte telomere length had a statistically discernible, but weak, association with mortality, but it did not predict survival as well as age or many other self-reported variables. Although telomere length may eventually help scientists understand aging, more powerful and more easily obtained tools are available for predicting survival.

Joint modeling of repeated multivariate cognitive measures and competing risks of dementia and death: a latent process and latent class approach.

Joint models initially dedicated to a single longitudinal marker and a single time-to-event need to be extended to account for the rich longitudinal data of cohort studies. Multiple causes of clinical progression are indeed usually observed, and multiple longitudinal markers are collected when the true latent trait of interest is hard to capture (e.g., quality of life, functional dependency, and cognitive level). These multivariate and longitudinal data also usually have nonstandard distributions (discrete, asymmetric, bounded, etc.). We propose a joint model based on a latent process and latent classes to analyze simultaneously such multiple longitudinal markers of different natures, and multiple causes of progression. A latent process model describes the latent trait of interest and links it to the observed longitudinal outcomes using flexible measurement models adapted to different types of data, and a latent class structure links the longitudinal and cause-specific survival models. The joint model is estimated in the maximum likelihood framework. A score test is developed to evaluate the assumption of conditional independence of the longitudinal markers and each cause of progression given the latent classes. In addition, individual dynamic cumulative incidences of each cause of progression based on the repeated marker data are derived. The methodology is validated in a simulation study and applied on real data about cognitive aging obtained from a large population-based study. The aim is to predict the risk of dementia by accounting for the competing death according to the profiles of semantic memory measured by two asymmetric psychometric tests.

Family history of myocardial infarction and cause-specific risk of myocardial infarction and venous thromboembolism: the Tromsø Study.

A family history of myocardial infarction (FHMI) has been shown to increase the risk of venous thromboembolism (VTE). The mechanism underlying the association remains unclear. Therefore, we aimed to determine the risks of MI and VTE by FHMI using a cause-specific model and to explore whether atherosclerotic risk factors could explain the association between FHMI and VTE in a population-based cohort. The study included 21 624 subjects recruited from the Tromsø Study in 1994 to 1995 and 2001 to 2002. Incident MI and VTE events were registered from date of enrollment to end of follow-up, December 31, 2010. There were 1311 MIs and 428 VTEs during a median follow-up of 15.8 years. FHMI was associated with a 52% increased risk of MI (adjusted hazard ratio, 1.52; 95% confidence interval, 1.35-1.70) and a 26% increased risk of VTE (adjusted hazard ratio, 1.26; 95% confidence interval, 1.02-1.55) in the cause-specific Cox model. Similar results were found using the traditional Cox model. The risk estimates by status of FHMI were highest for unprovoked deep vein thrombosis (adjusted hazard ratio, 1.69; 95% confidence interval, 1.12-2.56), and the risk increased with increasing number of affected relatives. Modifiable atherosclerotic risk factors slightly altered the association between FHMI and MI but had a negligible effect on the association between FHMI and VTE. FHMI was associated with increased risk of both MI and VTE in a cause-specific model. Apparently, the association between FHMI and VTE applied to unprovoked deep vein thrombosis and was not explained by modifiable atherosclerotic risk factors.

Added value of soluble tumor necrosis factor-α receptor 1 as a biomarker of ESRD risk in patients with type 1 diabetes.

Recent studies have suggested that circulating levels of the tumor necrosis factor-α receptor 1 (sTNFαR1) may be a useful predictor for the risk of end-stage renal disease (ESRD) in patients with diabetes. However, its potential utility as a biomarker has not been formally quantified. Circulating levels of sTNFαR1 were assessed in 429 patients with type 1 diabetes and overt nephropathy from the Finnish Diabetic Nephropathy (FinnDiane) cohort study. Predictors of incident ESRD over a median of 9.4 years of follow-up were determined by Cox regression and Fine-Gray competing risk analyses. The added value of sTNFαR1 was estimated via time-dependent receiver operating characteristic curves, net reclassification index (NRI), and integrated discrimination improvement (IDI) for survival data. A total of 130 individuals developed ESRD (28%; ESRD incidence rate of 3.4% per year). In cause-specific modeling, after adjusting for baseline renal status, predictors of increased incidence of ESRD in patients with overt nephropathy were an elevated HbA1c, shorter duration of diabetes, and circulating levels of sTNFαR1. Notably, sTNFαR1 outperformed estimated glomerular filtration rate in terms of R(2). Circulating levels of the sTNFαR1 also remained associated with ESRD after adjusting for the competing risk of death. A prediction model including sTNFαR1 (as a -0.5 fractional polynomial) was superior to a model without it, as demonstrated by better global fit, an increment of R(2), the C index, and area under the curve. Estimates of IDI and NRI(>0) were 0.22 (95% CI 0.16-0.28; P < 0.0001) and 0.98 (0.78-1.23; P < 0.0001), respectively. The median increment in the risk score after including sTNFαR1 in the prediction model was 0.18 (0.12-0.30; P < 0.0001). Circulating levels of sTNFαR1 are independently associated with the cumulative incidence of ESRD. This association is both significant and biologically plausible and appears to provide added value as a biomarker, based on the absolute values of NRI and IDI.

Competing risks of death in younger and older postmenopausal breast cancer patients

To show a new paradigm of simultaneously testing whether breast cancer therapies impact other causes of death. MA.14 allocated 667 postmenopausal women to 5 years of tamoxifen 20 mg/daily ± 2 years of octreotide 90 mg, given by depot intramuscular injections monthly. Event-free survival was the primary endpoint of MA.14; at median 7.9 years, the tamoxifen+octreotide and tamoxifen arms had similar event-free survival (P = 0.62). Overall survival was a secondary endpoint, and the two trial arms also had similar overall survival (P = 0.86). We used the median 9.8 years follow-up to examine by intention-to-treat, the multivariate time-to-breast cancer-specific (BrCa) and other cause (OC) mortality with log-normal survival analysis adjusted by treatment and stratification factors. We tested whether baseline factors including Insulin-like growth factor 1 (IGF1), IGF binding protein-3, C-peptide, body mass index, and 25-hydroxy vitamin D were associated with (1) all cause mortality, and if so and (2) cause-specific mortality. We also fit step-wise forward cause-specific adjusted models. The analyses were performed on 329 patients allocated tamoxifen and 329 allocated tamoxifen+octreotide. The median age of MA.14 patients was 60.1 years: 447 (82%) < 70 years and 120 (18%) ≥ 70 years. There were 170 deaths: 106 (62.3%) BrCa; 55 (32.4%) OC, of which 24 were other malignancies, 31 other causes of death; 9 (5.3%) patients with unknown cause of death were excluded from competing risk assessments. BrCa and OC deaths were not significantly different by treatment arm (P = 0.40): tamoxifen patients experienced 50 BrCa and 32 OC deaths, while tamoxifen + octreotide patients experienced 56 BrCa and 23 OC deaths. Proportionately more deaths (P = 0.004) were from BrCa for patients < 70 years, where 70% of deaths were due to BrCa, compared to 54% for those ≥ 70 years of age. The proportion of deaths from OC increased with increasing body mass index (BMI) (P = 0.02). Higher pathologic T and N were associated with more BrCa deaths (P < 0.0001 and 0.002, respectively). The cumulative hazard plot for BrCa and OC mortality indicated the concurrent accrual of both types of death throughout follow-up, that is the existence of competing risks of mortality. MA.14 therapy did not impact mortality (P = 0.77). Three baseline patient and tumor characteristics were differentially associated with cause of death: older patients experienced more OC (P = 0.01) mortality; patients with T1 tumors and hormone receptor positive tumors had less BrCa mortality (respectively, P = 0.01, P = 0.06). Additionally, step-wise cause-specific models indicated that patients with node negative disease experienced less BrCa mortality (P = 0.002); there was weak evidence that, lower C-peptide (P = 0.08) was associated with less BrCa mortality, while higher BMI (P = 0.01) was associated with worse OC mortality. We demonstrate here a new paradigm of simultaneous testing of therapeutics directed at multiple diseases for which postmenopausal women are concurrently at risk. Octreotide LAR did not significantly impact breast cancer or other cause mortality, although different baseline factors influenced type of death.

Parametric Cause-Specific Hazard Modeling with Nonproportional Covariate Effects: A Case Study Using Liver Transplant Data

We wanted to determine if parametric cause-specific hazard modeling, including log-linear generalizations of underlying parameters to incorporate covariate effects, would provide accurate representations, particularly with nonproportional hazards. Nonparametric cumulative hazard estimates were used for visual display. In the case study, the hazard rate of death-with-a-functioning-graft-due-to-infection following orthotopic liver transplantation (OLT) was modeled (N = 877, 82 such deaths). The three-parameter Makeham–Gompertz hazard, decreasing exponentially from time zero towards an asymptotic lower bound, and a more flexible, four-parameter mixture of generalized gamma functions (MGGF), increasing from time 0 towards a maximum value, then decreasing over time towards an asymptotic lower bound, were fitted. Both underlying hazards provided close fits, with a more accurate fit of MGGF during the first few months post-OLT. Parametric modeling of the important prognosticator donor age’s (≥ 60 vs. < 60yr) disappearing effect over time was achieved using two additional parameters in each case; a similar result was obtained using Cox’s model and a covariate by quadratic function of time interaction effect; however, Cox model fitted hazard ratios were overly inflated towards the end of the range of observed death times. In conclusion, the main lesson learned was the practicality in using a complete parametric modeling approach to better explain nonproportionality.

Change-point diagnostics in competing risks models: Two posterior predictive p-value approaches.

This paper presents a Bayesian diagnostic procedure for examining change-point assumption in the competing risks model framework. It considers the family of distributions arising from the cause-specific model as reported by Chiang (Introduction to stochastic processes in biostatistics. Wiley, New York, 1968) upon which change-points are added to accommodate possible distributional heterogeneity. Model departure, due to misspecification of change-points associated with either the overall survival distribution or cause-specific probabilities, is quantified in terms of a sequence of cumulative-sum statistics between each pair of adjacent change-points assumed. When assessing the asymptotic behavior of each sequence of cumulative-sum statistics using its posterior predictive p-values, see Rubin (Ann Stat 12:1151-1172, 1984) and partial posterior predictive p-values as reported by Bayarri and Berger (J Am Stat Assoc 95:1127-1142, 2000), we show that both types of p-values attain their greatest departure from 0.5 at the change-point that is missed in the assumed model, from which a diagnostic procedure is formalized. Statistical power of these two types of p-values is discussed.

Mortality in England and Wales attributable to any drinking, drinking above sensible limits and drinking above lowest-risk level.

To quantify mortality attributable to any alcohol consumption, and mortality attributable to consumption above different levels. We related all-cause mortality to alcohol consumption using cause-specific mortality models from a systematic review and using the distribution of alcohol consumption and causes of death by age and sex in England and Wales in 1997. We estimated the deaths and person-years of life lost to age 65 that were attributable: to any drinking; to drinking above the nadir (the level of alcohol consumption carrying the lowest risk); and to drinking more than the British Royal Colleges' recommended limits of 21 units/week in men and 14 units/week in women. Ischaemic heart disease deaths prevented by alcohol consumption (11 276 in men, 4050 in women) roughly balanced other deaths attributable to alcohol consumption (9246 in men, 4216 in women). Overall, 0.8% of all deaths in men were prevented by alcohol consumption (95% confidence interval, 0.2% to 1.3%), while 0.1% of all deaths in women were attributable to alcohol consumption (95% confidence interval, - 0.3% to 0.4%); 2.1% (1.9-2.3%) of all deaths in men and 0.8% (0.6-1.0%) of all deaths in women were attributable to drinking more than the recommended limits, while 2.8% and 1.2% of deaths, respectively, were attributable to drinking above the nadir. Of all person-years of life lost to age 65, 10.3% in men and 5.6% in women were attributable to any drinking; 8.5% and 4.0% were attributable to drinking above the recommended limits; and 12.6% and 6.0% were attributable to drinking above the nadir. Although overall mortality risks and benefits of alcohol consumption appear roughly equal, drinking above recommended limits remains responsible for many deaths and a large loss of person-years of life.

Changes over calendar time in the risk of specific first AIDS-defining events following HIV seroconversion, adjusting for competing risks.

Although studies have reported large reductions in the risks of AIDS and death since the introduction of potent anti-retroviral therapies, few have evaluated whether this has been similar for all AIDS-defining diseases. We wished to evaluate changes over time in the risk of specific AIDS-defining diseases, as first events, using data from individuals with known dates of HIV seroconversion. Using a competing risks proportional hazards model on pooled data from 20 cohorts (CASCADE), we evaluated time from HIV seroconversion to each first AIDS-defining disease (16 groups) and to death without AIDS for four calendar periods, adjusting for exposure category, age, sex, acute infection, and stratifying by cohort. We compared results to those obtained from a cause-specific hazards model. Of 6,941, 2,021 (29%) developed AIDS and 437 (6%) died without AIDS. The risk of AIDS or death remained constant to 1996 then reduced; relative hazard = 0.89 (95% CI: 0.77-1.03); 0.90 (95% CI: 0.81-1.01); and 0.32 (95% CI: 0.28-0.37) for 1979-1990, 1991-1993, and 1997-2001, respectively, compared to 1994-1996. Significant risk reductions in 1997-2001 were observed in all but two AIDS-defining groups and death without AIDS in a competing risks model (with similar results from a cause-specific model). There was significant heterogeneity in the risk reduction across events; from 96% for cryptosporidiosis, to 17% for death without AIDS (P < 0.0001). These findings suggest that studies reporting a stable trend for particular AIDS diseases over the period 1979-2001 may not have accounted for the competing risks among other events or lack the power to detect smaller trends.

Is the time from HIV seroconversion a determinant of the risk of AIDS after adjustment for updated CD4 cell counts?

To evaluate the effect of time from seroconversion to a given CD4 cell count on progression to AIDS after that count after adjusting for updated CD4 cell counts. Using pooled data from 19 seroconverter cohorts, we examined the association between the time from a CD4 <500 cells/mm 3, (<350, <200) to the first AIDS-defining event and time from seroconversion to that CD4 threshold. We adjusted for age, gender, exposure category, and HIV test interval in Cox models stratified by cohort. We estimated the residual effect of time from seroconversion, adjusting for updated CD4 cell counts. A cause-specific competing-risks model was then used to evaluate this residual effect on progression to each AIDS-defining disease. Analyses were censored on December 31, 1995. Of 3825, 3006, and 1804 individuals reaching CD4 thresholds of 500, 350, and 200, respectively, 1274, 1192, and 985, respectively, developed AIDS. We found a significant effect of time from seroconversion on the risk of AIDS even after adjusting for updated CD4 counts. For individuals reaching a CD4 threshold of 350 cells/mm 3, a 1-year increase from seroconversion was associated with an increase in risk of AIDS of 6% (3%-9%) ( p =.01). This effect appeared to be nonlinear. In the first 4 years, a 1-year increase from seroconversion was associated with an 11% increase in the risk of AIDS, but there was no apparent increase in risk after 4 years. The residual effect of time from seroconversion was significantly heterogeneous ( p =.002), with respect to the risk of individual AIDS-defining diseases. Findings were similar for CD4 thresholds of 500 and 200 cells/mm 3, respectively. We found a small, statistically significant, residual effect of time from seroconversion on the risk of AIDS. In practical terms, when considering an infected individual's risk of AIDS from a given CD4 cell count, there is little to be gained from knowing the time of seroconversion. However, this effect differs significantly among specific AIDS-defining diseases.

Association of elevated serum PO(4), Ca x PO(4) product, and parathyroid hormone with cardiac mortality risk in chronic hemodialysis patients.

Hyperphosphatemia is highly prevalent among patients with end-stage renal disease (ESRD) and is associated with increased mortality risk in hemodialysis (HD) patients. The mechanism through which this mortality risk is mediated is unclear. Data from two national random samples of HD patients (n = 12,833) was used to test the hypothesis that elevated serum PO(4) contributes mainly to cardiac causes of death. During a 2-yr follow-up, the cause-specific relative risk (RR) of death for patients was analyzed separately for several categories of cause of death, including coronary artery disease (CAD), sudden death, and other cardiac causes, cerebrovascular and infection. Cox regression models were fit for each of the eight cause of death categories, adjusting for patient demographics and non-cardiovascular comorbid conditions. Time at risk for each cause-specific model was censored at death that resulted from any of the other causes. Higher mortality risk was seen for patients in the high PO(4) group (>6.5mg/dl) compared with the lower PO(4) group (< or =6.5mg/dl) for death resulting from CAD (RR 1.41; P < 0.0005), sudden death (RR 1.20; P < 0.01), infection (RR 1.20; P < 0.05), and unknown causes (RR 1.25; P < 0.05). Patients in the high PO(4) group also had non-significantly increased RR of death from other cardiac and cerebrovascular causes of death. The RR of sudden death was also strongly associated with elevated Ca x PO(4) product (RR 1.07 per 10 mg(2)/dl(2); P < 0.005) and serum parathyroid hormone levels greater than 495 pg/ml (RR 1.25; P < 0.05). This study identifies strong relationships between elevated serum PO(4), Ca x PO(4) product, and parathyroid hormone and cardiac causes of death in HD patients, especially deaths resulting from CAD and sudden death. More vigorous measures to reduce the prevalence of these factors in HD patients may result in improved survival.

Methods for conducting sensitivity analysis of trials with potentially nonignorable competing causes of censoring.

We consider inference for the treatment-arm mean difference of an outcome that would have been measured at the end of a randomized follow-up study if, during the course of the study, patients had not initiated a nonrandomized therapy or dropped out. We argue that the treatment-arm mean difference is not identified unless unverifiable assumptions are made. We describe identifying assumptions that are tantamount to postulating relationships between the components of a pattern-mixture model but that can also be interpreted as imposing restrictions on the cause-specific censoring probabilities of a selection model. We then argue that, although sufficient for identification, these assumptions are insufficient for inference due to the curse of dimensionality. We propose reducing dimensionality by specifying semiparametric cause-specific selection models. These models are useful for conducting a sensitivity analysis to examine how inference for the treatment-arm mean difference changes as one varies the magnitude of the cause-specific selection bias over a plausible range. We provide methodology for conducting such sensitivity analysis and illustrate our methods with an analysis of data from the AIDS Clinical Trial Group (ACTG) study 002.

Postmenopausal estrogen use and mortality. Results from a prospective study in a defined, homogeneous community.

The authors studied the association between postmenopausal estrogen use and mortality from cardiovascular disease, coronary heart disease, cancer, and all causes in a cohort of 1,868 women aged 50-79 years residing in a planned community. After 12 years, the age-adjusted all-cause mortality rate was lower in the 734 postmenopausal estrogen users (14.9/100 women) compared with the 1,134 nonusers (21.5/100 women) (relative risk (RR) = 0.69, 95% confidence interval (Cl) 0.55-0.87). After adjustment for age, systolic blood pressure, social class, fasting plasma cholesterol, fasting plasma glucose, Quetelet index (weight (lbs)/height (in)2 x 100), and cigarette smoking by the Cox model, the relative risk increased to 0.79 (95% Cl 0.62-1.01). Because a postmenopausal estrogen-smoking interaction term was significant (p = 0.025), separate Cox models were run for never, past, and current smokers. In never and current smokers, estrogen was protective for all-cause mortality, with relative risks of 0.67 (95% Cl 0.45-0.99) and 0.62 (95% Cl 0.39-0.98), respectively. However, past smokers were not protected by postmenopausal estrogen use (RR = 1.32, 95% Cl 0.84-2.08). Cause-specific models revealed differences in the association of postmenopausal estrogen use with cardiovascular disease mortality and coronary heart disease mortality that were dependent on smoking status. Postmenopausal estrogen use was strongly protective in current smokers but was associated with increased risk in past smokers. As expected, cancer mortality was increased in smokers. The confidence intervals for the relative risk estimate of postmenopausal estrogen use for cancer mortality in each smoking category included one. Finally, a separate analysis of subsequent three-year mortality in women surviving the first nine years of follow-up revealed reduced death rates only for women using estrogen at both baseline and nine years of follow-up, suggesting both a conservative bias in our data introduced by the large reduction in postmenopausal estrogen use during the study period and the possibility of a stronger protective effect for recent postmenopausal estrogen use.