Chapter 2 Chapter 2 Ethics in Epidemiology: Common Misconceptions, Paradoxes, and Unresolved Questions Affiliation Steven S. Coughlin Steven S. Coughlin , Ph.D. CopyRight https://doi.org/10.2105/9780875531939ch02 Published Online: August 09, 2012 © American Public Health Association
BACKGROUND Hepatitis C is transmitted by transfusion of unscreened blood, through injecting drugs, from mother-to-child and, on occasion, sexually. Transmission generally requires that the infector is hepatitis C virus (HCV) RNA positive, a 'carrier'. About three-quarters of injectors who are hepatitis C antibody positive are HCV-RNA positive and so infectious to others. Incubation periods from HCV infection to cirrhosis and hepatocellular carcinoma are even longer than from HIV infection to AIDS, being counted in decades; they depend on age, gender, alcohol consumption and co-infection with other viruses. We identify 25 data sources that are available, or required, for projecting the severe sequelae of the injection-related hepatitis C epidemic. DATA SOURCES Three data sources relate to hepatitis C diagnosis: register of confirmed HCV infections (with initial of first name + soundex of surname + date of birth + gender = master index, exposure category, year of starting to inject, and region); surveys of HCV test-uptake by injectors and others; documentation of pregnancy and its outcome in HCV-infected women (injectors and others). Four data sources relate to HCV prevalence and incidence among injectors and others: anonymous testing for HCV antibodies in blood or saliva (for sentinel groups ranging from new blood donors, pregnant women, patients awaiting kidney transplantation, non-injector prisoners, health-care workers, non-injector heterosexuals attending genitourinary medicine clinics; to injectors in the community, at drug treatment centres or in prison); historical data on HCV prevalence in injectors; HCV incidence studies in injectors; and uptake of harm reduction measures--frequency of sharing and methadone substitution--by injectors. Key reporting problems in HCV incidence studies, which inhibit checks on the convenient exponential assumption for time from start of injecting to hepatitis C infection, are discussed. Nine critical data sources are identified for monitoring the late sequelae of hepatitis C carriage, its investigation and treatment: linkage surveillance, for example by master index, to identify deaths, hospitalisations or cancer registrations among confirmed HCV infections; surveys of HCV status among patients who undergo liver biopsy, are newly diagnosed with cirrhosis or are newly diagnosed with liver cancer; surveys of liver-biopsy rate in HCV-infected injectors and others; uptake and outcome of interferon + ribavirin in the treatment of hepatitis C carriers; cohort studies of HCV progression; sample surveys of genotype in HCV-infected injectors, and others; acute hepatitis B infections and uptake of hepatitis B immunisation by injectors; liver transplantation in HCV-infected patients; and hepatitis C-status and other risk factors in deaths from cirrhosis or liver cancer, to determine whether they are HCV and injector-related. Finally, nine critical data sources are identified for quantitative understanding of the underlying injector epidemic: drug misuse databases plus capture-recapture methods to assess number of injectors, drug-related deaths by region to assess injector numbers; number of HIV-infected injectors; HIV progression in injectors; overdose and other causes of death in injectors; expert opinion on injector incidence historically, plus survey information on age-distribution at initiation and duration of injector careers; injector incidence historically inferred from hepatitis C infected blood donors; age-distribution of current injectors and at initiation, as a check on the assumptions made in stochastic simulation about injector incidence and 'outcidence' from injecting historically; mortality of former injectors; and general population or other survey ratios of surviving ever-injectors to injectors in the last 5 years, last year and currently, as a check on simulations. RECOMMENDATIONS We recommend a common HCV diagnosis report form to improve ascertainment of risk-factor information, especially year of starting to inject--which is a key date epidemiologically. We also recommend updated surveys of current and former injectors' HCV-test uptake, or a denominator study that registers master index and risk factor information for all HCV testees. We recommend that injector surveys ask about typical frequency of needle sharing per 4 weeks in three distinct periods this year, last year and in the first year of injecting. We also recommend the location of stored historical samples from injectors to be tested retrospectively and anonymously for HCV antibodies. We recommend immediate attention to the uptake of, and response to, combination treatment by hepatitis C carriers who are former or recovering injectors. We rec
BACKGROUND This study was devised to determine the prevalence of urinary symptoms among men living in the Australian cities of Melbourne, Sydney or Perth, and to identify factors associated with the presence of moderate-to-severe urinary symptoms. METHODS The study comprised a population-based sample of 1,216 men, aged 40-69 years, whose names were obtained through electoral rolls and who participated as controls in a case-control study of risk factors for prostate cancer. As part of a structured face-to-face interview, the men completed the International Prostate Symptom Score (IPSS). Men with moderate (IPSS = 8-19) or severe (IPSS > or = 20) urinary symptoms were compared with those with mild or no symptoms (IPSS < 8) using unconditional logistic regression. RESULTS The age-specific prevalence of moderate-to-severe urinary symptoms (IPSS > or = 8) in men aged 40-49, 50-59, 60-69 years was 16%, 23% and 28%, respectively. Compared with men with no or mild urinary symptoms (IPSS < 8), men with moderate-to-severe symptoms were more likely to report not currently living as married [odds ratio (OR) = 1.5; 95% confidence interval (CI) 1.1-2.0] and being circumcised (OR = 1.5; 95% Cl 1.2-2.0). The increased likelihood associated with drinking an average of > 60 g day(-1) of alcohol in the 2 years before interview was of marginal statistical significance (OR = 1.6; 1.0-2.6). There were no significant differences between men with IPSS > or = 8 and those with IPSS < 8 with respect to body mass index, education level, having had a vasectomy, or cigarette smoking. CONCLUSION Among Australian men, being circumcised, or not currently living as married, were associated with increased prevalence of urinary symptoms.
BACKGROUND:Relative survival is a method of analysis of failure-time data used to estimate the net survival. Cancer registries frequently use this method. The main regressive models are the Hakulinen and Tenkanen model, and the Esteve et al. model, which are easily used in practice thanks to their specific software (SURV and RELSURV, respectively). An assessment of the behaviour of the models is made, with the aim of giving advice for users of lifetime data in practice.METHODS:Simulations were done by respecting, then violating, the basic hypothesis supporting the theoretical foundation of these two proportional hazard models (independence of the death and censor process, proportionality of risks). For each simulation, 100 files of either 100, 1,000, or 10,000 individuals were generated to assess the behaviour of the model.RESULTS:Moderate censor rates, with or without proportionality assumption, lead to the use of the Hakulinen and Tenkanen model, especially for studies with little information. Non-proportionality of risks in the Hakulinen and Tenkanen model could be tested and analysed. If assumptions underlying the models are respected, the Esteve et al. model seems to be more precise.DISCUSSION:The choice of a model in practice depends on its performance, and on the user's knowledge of statistics and computer science. Non-proportionality of risks is common in cancer registries. In theory, non-proportionality of risks could be taken into account for both relative survival models but, for the moment, it is feasible in routine only for the Hakulinen and Tenkanen model. Characteristics of the software should also be taken into account for routine relative survival analyses.
BACKGROUND:It has been suggested that prolonged exposure to sunlight may induce systemic or local immune alterations, which may facilitate the development of skin cancer and, perhaps, non-Hodgkin's lymphona. The effects of prolonged sunlight exposure on peripheral blood cells were studied.METHODS:Leukocytes and lymphocyte subpopulations of 12 volunteers aged 10-45 were investigated before and after a 3-week summer holiday in seaside resorts in Greece. Lymphocyte phenotypes were estimated using monoclonal antibodies and flow cytometry.RESULTS:There were no significant differences with respect to total numbers of T cells, T-helper/inducer, T-suppressor/cytotoxic, B cells or HLA-Dr+ cells. However, we have found evidence of lymphocyte stimulation, reflected in an increase in cells expressing the interleukin-2 receptor (IL-2R) and, more specifically, an increase in the T cells expressing IL-2R and HLA-Dr antigens. An increase in natural killer cells has also been noticed.CONCLUSIONS:These findings suggest that prolonged intense exposure to sunlight may be associated with immunostimulation, rather than immunosuppression.
BACKGROUND:Epidemiological studies of the effects of hormone replacement therapy (HRT) often rely on exposure data and information on past health from self-administered questionnaires. The accuracy with which women report current use of HRT and the specific preparation in use is not known. This study aims to compare aspects of self-reported use of HRT and treatment for various conditions with data from general practice prescription records.METHODS:Reported questionnaire data on use of HRT were compared with those on the general practice prescription record for 570 women participating in the Million Women Study from two general practices in the UK.RESULTS:There was excellent agreement between data from the self-administered questionnaire and the prescription record: 96% agreement (kappa = 0.91) for current use of HRT, 95% agreement (kappa = 0.90) for any use of HRT during the period covered by the prescription record, and 97% agreement (kappa = 0.95) among current users for whether the HRT preparation contained oestrogen alone, combined oestrogen/progestogen, or some other constituents. Among former HRT users who provided questionnaire information on the preparation they used most recently, there was 69% agreement on the proprietary preparation used and 97% agreement (kappa = 0.93) on the hormonal constituents used. Agreement between reported treatment for various conditions and the presence of a prescription appropriate for that condition ranged from 89-99% (kappa 0.53-0.92), and was highest for thyroid disease and asthma.CONCLUSION:Important aspects of use of HRT, such as type of preparation currently being used, are reported very reliably by women completing a self-administered questionnaire.
P-values are a practical success but a critical failure. Scientists the world over use them, but scarcely a statistician can be found to defend them. Bayesians in particular find them ridiculous, but even the modern frequentist has little time for them. In this essay, I consider what, if anything, might be said in their favour.
BACKGROUND:European guidelines for breast-cancer screening recommend an integrated approach of mammography screening with subsequent assessment and biopsy, if required, in one screening unit under permanent quality control, for which target values are released. Although the calculation of the respective rates (e.g. for participation, assessment, biopsy, or cancer detection) appears trivial, the statistical assessment of their compatibility with the target values is less obvious. This is especially true if subjects with a positive diagnostic result leave the screening-assessment chain prematurely, and information about further diagnostic results outside the organised screening is lacking.METHOD:Statistical models for the basic situation, in which complete information about the screening and assessment outcome is available, as well as for when information is incomplete, are presented. The statistical methods for obtaining the confidence limits, statistical tests and sample sizes needed to obtain a desired power of tests for the process parameters of interest are also given.RESULTS:The sample-size calculations indicate that large numbers of enrolled subjects are required to obtain reasonably narrow confidence limits, and that incomplete information about the outcome of diagnostic procedures among screening positives considerably worsens the feasibility of quality control.CONCLUSIONS:Although the methodology is specified for breast-cancer screening, it should be adaptable easily to other screening issues.
BACKGROUND:Data visualisation has become an integral part of statistical modelling.METHODS:We present visualisation methods for preliminary exploration of time-series data, and graphical diagnostic methods for modelling relationships between time-series data in medicine. We use exploratory graphical methods to better understand the relationship between a time-series reponse and a number of potential covariates. Graphical methods are also used to examine any remaining information in the residuals from these models.RESULTS:We applied exploratory graphical methods to a time-series data set consisting of daily counts of hospital admissions for asthma, and pollution and climatic variables. We provide an overview of the most recent and widely applicable data-visualisation methods for portraying and analysing epidemiological time series.DISCUSSION:Exploratory graphical analysis allows insight into the underlying structure of observations in a data set, and graphical methods for diagnostic purposes after model-fitting provide insight into the fitted model and its inadequacies.
BACKGROUNDCase-control research is often exploratory; to determine factors that increase risk. Often, regression methods are used to determine combinations of risk factors that predispose to excess risk. Recently, tree-based methods have also been proposed. Both have limitations. An alternative approach is suggested, based on a search algorithm to identify at-risk subgroups.METHODSStatistical methods to determine and visualise at-risk sub-groups in case-control studies are presented. The method of determining sub-groups--search partition analysis (SPAN)--searches among different Boolean combinations of risk factors. Sub-groups that have been identified are visualised by scaled rectangle diagrams. These show the size of sub-groups and the extent to which they overlap.RESULTSTheory is presented for applying the method to case-control data. The methods are illustrated by analysis of three case-control studies: one on sudden infant death syndrome, a second on heart disease and a third on child pedestrian injuries.CONCLUSIONSThe methods provide a useful alternative to regression and tree-based analysis. They demarcate subgroups that, in the three examples, are easy to interpret and would not have been found by other methods. Scaled rectangle diagrams are a useful way to visualise the results.
BACKGROUND Different approaches have been proposed to investigate latency in epidemiologic studies where detailed exposure histories are available. METHODS We demonstrate the application of a flexible, yet parsimonious, spline function model to investigate latency patterns for radon progeny exposure and lung cancer in the Colorado Plateau uranium miners cohort. The model extends a previously proposed bilinear model. RESULTS The excess relative risk (ERR) reached a maximum of 0.6 per 100 working level months, for exposures received 14 years previously. The ERR then declined, and was estimated to approach zero for exposures received 35 years and more in the past. The point-wise 95% confidence intervals supported ERRs > 0 for the period 9-32 years before the event. The estimated latency curve was homogeneous across categories of attained age, duration of exposure, rate of exposure, and smoking. CONCLUSIONS The proposed spline model is a flexible tool for latency analyses, and extends previously used methods.
BACKGROUND:Markov and semi-Markov models are increasingly used in clinical and public health epidemiology to represent disease processes. We present a Markov model of events following lung transplantation as a case study in clinical epidemiology.METHODS:A five-state discrete-time Markov model with two-way transitions between acute event states is applied to the analysis of 356 lung transplant patients. A two-state continuous time Markov model for chronic disease onset is fitted. Values of transition parameters are estimated by maximum likelihood using numerical methods.RESULTS:Accurate estimates of acute and chonic event rates, and survival probabilities are calculated from transition probabilities. Costs attributed to different acute and chronic states are calculated.CONCLUSIONS:Transition models provide a useful and flexible representation of acute and chronic events and can be used to explore the economic impact of changes in therapy.
BACKGROUND:In Part 2, we illustrate how available data can be used to obtain preliminary estimates for Scotland of prevalent injection-related hepatitis C carriers and of maternally hepatitis C virus (HCV)-infected infants. Novel approaches to reducing uncertainty about the number of Scotland's HCV infected children of injector parents are discussed in brief. Three approaches, one direct and two indirect, to estimating the number of current and ever-injectors are presented for England and Wales.METHODS:Diagnosed HCV infections in injectors and HCV test uptake by current injectors are combined with survey estimates for the ratio of ever-injectors to current injectors to estimate prevalent injection-related hepatitis C carriers. Household surveys give direct but potentially biased estimates of the number of current and ever-injectors. Indirect estimates make use of hepatitis C diagnoses in injectors, HCV prevalence and test-uptake by injectors, or exploit international comparisons. We comment on key reporting problems that inhibit synthesis of HCV progression studies; and suggest how to derive preliminary gender-and-age specific progression rates to liver cirrhosis for use in projections.RESULTS:Preliminary estimates for Scotland of prevalent injection-related hepatitis C carriers are: central estimate 39,000, inner uncertainty 16,000-59,000; of maternally hepatitis C virus (HCV)-infected infants central estimate 260, uncertainty 110-1100; and for England and Wales estimates of the number of prevalent ever-injectors are central estimate 360,000, uncertainty 240,000-835,000. Both hepatitis C prevalence in injectors and estimated numbers of current injectors are similar in Australia, and England and Wales (but not so for Scotland), Australian work on projections of severe HCV sequelae from hepatitis C infections may therefore be a suitable starting point for projections for England and Wales. Australia anticipates a doubling in the number of persons living with hepatitis C cirrhosis from 8500 in 1997 to over 17,000 in 2010.DISCUSSION:Australian projections of severe HCV sequelae used progression rates that, for simplicity, were independent of gender and of age at HCV infection. Faster HCV progression for males, and their higher injector prevalence, means that the impact of HCV infection on, for example, liver cancer may be evident to a greater extent and earlier in males.
BACKGROUND:SF-12 is a generic short form health survey, developed in the USA from the original SF-36. It produces two summary measures evaluating physical and mental self-perceived health that are interchangeable with those from the SF-36. SF-12 has been successfully tested in nine Western European countries on large samples of the general population, where it has proved its brevity, comprehensiveness, reliability, validity and cross-cultural applicability. The present analysis directly assesses the SF-12 for the first time in various Italian settings, including the general population and specific patient groups.METHODS:Data for this report were collected from five different samples; in four of them the SF-12 was used as a 'stand-alone' instrument, while in the other one (used as the reference) it was embedded in the SF-36. Descriptive statistics, Spearman's correlation coefficients, confirmatory factor analysis, ordinal uni- and multi-variate least squares regression model and covariance analysis were used to evaluate the summary measures in each sample, and across relevant subgroups. Studies were ordered according to the expected deviance, from the 'normal' health status of the reference group to the sample with the expected highest level of illness.RESULTS:Overall, more than 11,000 subjects were evaluated. Response rates ranged from 63 to 100%, while missing items accounted only for 0.2-8.2% of all items. Uni- and multi-variate analyses showed a positive association between both physical component summary (PCS) and mental component summary (MCS) scores and their respective items in all examined samples. MCS scores were fairly similar across all samples, with the only exception being patients recently discharged from hospital, whose subjective mental health perception was higher than expected and the highest of all (52.2). Finally, we found a substantial impact of ageing on physical health perception, while the MCS was shown to be less sensitive to the age effect.CONCLUSIONS:This analysis shows that the SF-12 has good validity, while some issues related to its most appropriate mode of administration and target groups might require further attention.
BACKGROUND:Estimations of mean sojourn time (MST) and sensitivity (S) in disease screening have been previously calculated from case-control data, using simple models which did not include covariates. Many studies have shown an effect of mammographic parenchymal pattern (MPP) on breast-cancer risk and tumour histology. We have expanded previous models on these to estimate MST and S with the effects of MPP as a covariate.METHODS:Data were from a nested case-control study within the East Anglian screening programme, with 875 cases and 2,601 controls. Estimates of disease progression and screening parameters were based on conditional likelihood calculation, using a Markov process model. Ninety-five per cent confidence intervals (CI) were calculated using the profile likelihood wherever possible and using a numerical estimate of the information matrix or the area under the likelihood curve where necessary.RESULTS:We obtained estimates of the incidence of preclinical disease, rate of transition from preclinical to clinical and screening sensitivity, and evaluated the association of these parameters with mammographic parenchymal pattern. A higher incidence of preclinical disease was found for high-risk MPP [relative incidence = 1.62 (95% CI: 0.89; 2.73)]. However, no difference in progression rate from preclinical to clinical disease between different MPP was found. Dense MPPs were associated with decreased sensitivity [relative sensitivity = 0.24 (95% CI: 0.06; 15)]. Wide CIs were found, probably being a consequence of the relative sparsity of interval cancer data.DISCUSSION:It is possible to estimate multiple parameters of disease progression and screening quality from case-control data. The reduction in sensitivity of the screening process associated with high-risk patterns presented here, could be of paramount interest for proposing new screening strategies, such as possible additional screening tools.
