Journal of epidemiology and biostatistics最新文献

Background: It is generally believed that dental caries is an infectious disease. The occurrence of dental caries is affected by a variety of determinants. In order to estimate the precise extent of the relation between specific determinants and the outcome phenomenon (i.e. the occurrence of dental caries), a coherent disease model is required. This model should also permit multivariate analysis to control for confounders and interactions. Only with such a disease model will it be possible to investigate the relation between the occurrence of a determinant and dental caries, and to estimate the extent of this relation. The known causal models for the explanation of dental caries do not fully meet these requirements.

Method: Rothman's 'sufficient cause' model has been used as a starting point for the development of a new coherent disease model, to explain the occurrence of dental caries and allow multivariate analysis.

Results: The sufficient cause for dental caries comprises three component causes: sufficient microorganisms with cariogenic potential, easily fermentable carbohydrates and teeth. Whether dental caries actually occurs also depends on the influence of independent risk factors that interact with the component causes in a protective, as well as in a risk-increasing manner. These independent risk factors are saliva, fluoride, oral hygiene and diet.

Conclusions: The 'sufficient cause' model for dental caries is a biological model in which distinction between protective and risk-increasing factors has been made, and interaction between factors has been described. With this model, it will now be possible to assess the extent of the relationship between a determinant and dental caries (the outcome phenomenon) using multivariate techniques.

Background: The Hosmer-Lemeshow test, used extensively to assess the fit of the logistic regression model, is performed by several statistical packages. Recent studies have shown some problems in the use of this test when ties are present. These problems were attributed merely to the test implementation.

Methods: We analysed the order of the observations as an alternative explanation of the problem of ties. Using a data-set of 1393 intensive care unit (ICU) patients we performed the Hosmer-Lemeshow test with all possible subjects dispositions.

Results: We obtained about one million different P values, ranging from 0.01 to 0.95.

Discussion: It is already known that when the Hosmer-Lemeshow goodness-of-fit test is performed with a number of covariate patterns lower than the number of subjects, its result may be inaccurate. We showed that the extent of this problem could be relevant under particular conditions. We also suggest a strategy for estimating the extent of the problem and subsequent interpretation.

Background: 'Life expectancy' (LE) is a health-status indicator in widespread use. However, LE is an index of central location, but not of dispersion. It cannot describe inter-individual variation in the life-span. This author proposes using the Gini coefficient, a summary index of the Lorenz curve, for characterising life-span variability. Like the LE, the proposed index is also based on a lifetable.

Method: The method is a nonparametric approach that does not make specific assumptions about mortality rates.

Results: The author uses vital statistics from Taiwan as a demonstration and finds that the method provides information crucial to the understanding of the epidemiologic transitions of the past 20 years (Gini decreases from 0.1320 to 0.1130). It also quantifies the impacts of elimination of some selected causes of death in Taiwan.

Conclusions: It is recommended that Gini be routinely compiled in official vital statistics, along with the LE.

During the 20th century alcohol and alcoholism have emerged as a problem with global health implications. In Westernized countries > or = 50% of adults can be classified as alcohol consumers. For most people, drinking is a safe, pleasurable experience with minimal health consequences. However, about 10% of alcohol consumers will at some time experience serious health problems related to their drinking habit. Persons at risk of drinking problems cannot be reliably identified in the population--a fertile area for additional research. At present, the World Health Organization estimates that > 15 million people are disabled as a result of alcohol use, making it the fourth leading cause of worldwide disability. The challenge for the 21st century is to reduce the impact of alcohol-related disease by measures including: * Identification of high risk individuals. * Social control. * More effective treatment modalities for people addicted to alcohol.

Background: AIDS represents an important public health problem in Italy. Long-term health care policy planning requires knowledge about the variability of the risk of mortality. The AIDS Surveillance Registry (RAIDS), maintained by the AIDS Operational Centre (COA) of the National Health Institute of Italy, provides valuable information to study the determinants of survival after diagnosis with AIDS. This study aimed to estimate the trends among people infected by the human immunodeficiency virus (HIV) through blood-related products.

Methods: Study subjects were 595 persons with AIDS whose sole ascertained risk factors were either blood transfusions or plasma-concentrate infusions, diagnosed from the beginning of the epidemic in 1985 through June 1995 and reported to RAIDS by the end of June 1996. The Kaplan-Meier technique was used to estimate the survival distribution; log-rank and Wilcoxon tests were both performed to assess the effects of demographic and clinical factors. Cox proportional hazards models were used to identify those factors independently and significantly associated with death: model building and fitting were performed in a stepwise fashion, by using the score and martingale residuals, based on a new class of graphical and numerical methods developed recently for checking the assumptions underlying the model.

Results and conclusions: In Italy the median survival time for AIDS patients infected by contaminated blood, was estimated to be 12.7 months. In univariate analyses it was found that women, younger patients (age < 35) and those diagnosed more recently with a higher value of CD4 cell counts (>37 cells m(-3)) have better survival. Patients diagnosed with AIDS-associated neurological disease (neuro-AIDS), or lymphoma, had a median survival significantly shorter. Patients diagnosed in the south of Italy tend to have a survival time shorter than patients diagnosed in the north. In a multivariate time-dependent regression analysis, only type of AIDS indicator disease, age and calendar time of diagnosis proved to be significant prognostic factors. It was not possible to estimate the effect of risk category (haemophiliacs versus transfused) due to the lack of proportionality in the estimated hazard. In conclusion, survival time is found to improve over time, indicating a likely positive effect of better care in treating HIV and AIDS patients.

Thesis: The UK Government Statistical Service reports the percentage of elective 'admissions' that took place in England within 3 months of a patient being added to NHS waiting lists. This percentage is calculated from cross-sectional data using the total number of elective episodes within a specified calendar period as denominator and the number of these enrolled on the waiting list less than 3 months previously as numerator. This approach assumes that NHS waiting lists are closed and stationary populations, and has been widely used by government and non-government researchers in the UK and elsewhere.

Antithesis: Little attention has been given to the bias introduced when waiting lists are neither stationary nor closed. This paper identifies four groups of patients which are excluded from the denominator used by the Government Statistical Service and criticises the established method of ignoring left and right censored observations.

Synthesis: We describe two alternative formulae that would give the same results as the Government Statistical Service method if waiting lists were closed and stationary, but that also give unbiased results when waiting lists are open and non-stationary. They require a limited amount of additional cross-sectional data to produce upper and lower estimates of the cumulative likelihood of admission among those listed. We recommend the production of unbiased estimates by applying period life-table techniques to a complete and consistent set of 'times since enrolment'.

Background: Analysing the geographical variation of cancer incidence is an important issue in epidemiological research. It might suggest new aetiologic hypotheses, provide guidelines for the design of new surveys and give ideas for preventive campaigns.

Methods: Four different methods for estimating the variation of cancer risks between small areas and three homogeneity tests were evaluated by simulation. In three of the methods the systematic variation of the relative risks (RR) was estimated by subtracting the expected Poisson variation from the total variation. The last method assumes that RR are gamma distributed and the maximum likelihood estimate (MLH) of the systematic variation parameter is calculated. A likelihood ratio test (LRT) of heterogeneity of RR based on this method is also evaluated, and compared with an ordinary chi2 test and the Potthoff and Whittinghill test (P&W).

Results: For most of the simulated data-sets, the estimates obtained by MLH are most precise, even if the assumption of gamma distribution of RR is violated. The LRT and P&W tests of homogeneity are also shown to perform better than the chi2 test. Most of the real cancer data-sets exhibited at least some geographical variation. Cancer of the lung, melanoma and other skin cancers, and cancers of the urinary bladder, pancreas and stomach, have the highest systematic variation.

Discussion: The study suggests that likelihood-based approaches are suitable, both for estimating the variation between small areas and for testing the null hypothesis of equal RR. Such geographical analyses might suggest new aetiological hypothesis.

Background: Although interpretation of age-period-cohort analyses is complicated by the non-identifiability of maximum likelihood estimates, changes in the slope of the birth-cohort effect curve are identifiable and have potential aetiologic significance.

Methods: A nonparametric test for a change in the slope of the birth-cohort trend has been developed. The test is a generalisation of the sign test and is based on permutational distributions. A method for identifying interactions between age and calendar-period effects is also presented.

Results: The nonparametric method is shown to be powerful in detecting changes in the slope of the birth-cohort trend, although its power can be reduced considerably by calendar-period patterns of risk. The method identifies a previously unidentified decrease in the birth-cohort risk of lung-cancer mortality from 1912 to 1919, which appears to reflect a reduction in the initiation of smoking by young men at the beginning of the Great Depression (1930s). The method also detects an interaction between age and calendar period in leukemia mortality rates, reflecting the better response of children to chemotherapy.

Conclusion: The proposed nonparametric method provides a data analytic approach, which is a useful adjunct to log-linear Poisson analysis of age-period-cohort models, either in the initial model building stage, or in the final interpretation stage.

Background: The methods used in epidemiological studies to assess exposure are often affected by a conspicuous amount of measurement error. Exposure-measurement error is recognised to cause attenuation in the association between exposure and disease. Among different possible approaches, the validity coefficient of a measurement can be estimated by a comparison of three types of measurements, using either structural equation models or factor analysis (the triads method). These approaches assume that the measurements are linearly related to true intake and have independent random errors.

Methods: In this paper we present an estimator of the variance of the estimated validity coefficient to compute the associated confidence intervals. Standard error for the validity coefficient allows the efficiency of validation studies to be evaluated. Our work was motivated by the fact that existing software does not provide correct standard errors for the estimated validity coefficient. The approach is illustrated using selected examples from dietary validation studies.

Results: The accuracy of our formula is evaluated by comparison with the results of a simulation study, which shows that our variance estimator provides good results for sample sizes of at least n = 100 and when the expected value of the validity coefficient is not too close to 1.0, independent of the sample size. Our estimator formula performs better than either a naïve approach, that computes the standard error for a validity coefficient as if it is a straightforward correlation coefficient, or the SAS-CALIS procedure, which uses a maximum likelihood method.

Conclusions: In evaluating the validity of the type of measurement chosen to assess exposure in an epidemiological study, it is important to provide an estimate of the precision of the validity coefficient of the measurement. Our variance estimator may help calculate sample size requirements for validation studies.