Accuracy of alternative approaches to capture-recapture estimates of disease frequency: internal validity analysis of data from five sources

Dependence-Robust Confidence Intervals for Capture-Recapture Surveys

Capture-recapture (CRC) surveys are used to estimate the size of a population whose members cannot be enumerated directly. CRC surveys have been used to estimate the number of Coronavirus Disease 2019 (COVID-19) infections, people who use drugs, sex workers, conflict casualties, and trafficking victims. When k-capture samples are obtained, counts of unit captures in subsets of samples are represented naturally by a 2 k contingency table in which one element-the number of individuals appearing in none of the samples-remains unobserved. In the absence of additional assumptions, the population size is not identifiable (i.e., point identified). Stringent assumptions about the dependence between samples are often used to achieve point identification. However, real-world CRC surveys often use convenience samples in which the assumed dependence cannot be guaranteed, and population size estimates under these assumptions may lack empirical credibility. In this work, we apply the theory of partial identification to show that weak assumptions or qualitative knowledge about the nature of dependence between samples can be used to characterize a nontrivial confidence set for the true population size. We construct confidence sets under bounds on pairwise capture probabilities using two methods: test inversion bootstrap confidence intervals and profile likelihood confidence intervals. Simulation results demonstrate well-calibrated confidence sets for each method. In an extensive real-world study, we apply the new methodology to the problem of using heterogeneous survey data to estimate the number of people who inject drugs in Brussels, Belgium.

Estimating the prevalence of problem drug use from drug-related mortality data

BACKGROUND AND AIMS

Indirect estimation methods are required for estimating the size of populations where only a proportion of individuals are observed directly, such as problem drug users (PDUs). Capture-recapture and multiplier methods are widely used, but have been criticized as subject to bias. We propose a new approach to estimating prevalence of PDU from numbers of fatal drug-related poisonings (fDRPs) using linked databases, addressing the key limitations of simplistic 'mortality multipliers'.

METHODS

Our approach requires linkage of data on a large cohort of known PDUs to mortality registers and summary information concerning additional fDRPs observed outside this cohort. We model fDRP rates among the cohort and assume that rates in unobserved PDUs are equal to rates in the cohort during periods out of treatment. Prevalence is estimated in a Bayesian statistical framework, in which we simultaneously fit regression models to fDRP rates and prevalence, allowing both to vary by demographic factors and the former also by treatment status.

RESULTS

We report a case study analysis, estimating the prevalence of opioid dependence in England in 2008/09, by gender, age group and geographical region. Overall prevalence was estimated as 0.82% (95% credible interval = 0.74-0.94%) of 15-64-year-olds, which is similar to a published estimate based on capture-recapture analysis.

CONCLUSIONS

Our modelling approach estimates prevalence from drug-related mortality data, while addressing the main limitations of simplistic multipliers. This offers an alternative approach for the common situation where available data sources do not meet the strong assumptions required for valid capture-recapture estimation. In a case study analysis, prevalence estimates based on our approach were surprisingly similar to existing capture-recapture estimates but, we argue, are based on a much more objective and justifiable modelling approach.

Prevalence, socio-demographic characteristics, and comorbid health conditions in pre-dialysis chronic kidney disease: results from the Manitoba chronic kidney disease cohort

Background

Chronic Kidney Disease (CKD) is common and its prevalence has increased steadily over several decades. Monitoring of rates and severity of CKD across populations is critical for policy development and resource planning. Administrative health data alone has insufficient sensitivity for this purpose, therefore utilizing population level laboratory data and novel methodology is required for population-based surveillance. The aims of this study include a) develop the Manitoba CKD Cohort, b) estimate CKD prevalence, c) identify individuals at high risk of progression to kidney failure and d) determine rates of comorbid health conditions.

Methods

Administrative health and laboratory data from April 1996 to March 2012 were linked from the data repository at the Manitoba Centre for Health Policy. Prevalence was estimated using three methods: a) all CKD cases in administrative and laboratory databases; b) all CKD cases captured only through the laboratory data; c) and the capture-recapture method. Patients were stratified by risk by estimated Glomerular Filtration Rate (eGFR) and albuminuria based on Kidney Disease Improving Global Outcomes (KDIGO) criteria. For comorbid health conditions, the counts were modelled using a Generalized Linear Model (GLM).

Results

The Manitoba CKD Cohort consisted of 55,876 people with CKD. Of these, 18,342 were identified using administrative health data, 27,393 with laboratory data, and 10,141 people were identified in both databases. The CKD prevalence was 5.6% using the standard definition, 10.6% using only people captured by the laboratory data and 10.6% using the capture-recapture method. Of the identified cases, 46% were at high risk of progression to end-stage kidney disease (ESKD), 41% were at low risk and 13% were not classified, due to unavailable laboratory data. High risk cases had a higher burden of comorbid conditions.

Conclusion

This study reports a novel methodology for population based CKD surveillance utilizing a combination of administrative health and laboratory data. High rates of CKD at risk of progression to ESKD have been identified with this approach. Given the high rates of comorbidity and associated healthcare costs, these data can be used to develop a targeted and comprehensive public health surveillance strategy that encompass a range of interrelated chronic diseases.

Electronic supplementary material

The online version of this article (10.1186/s12882-018-1058-3) contains supplementary material, which is available to authorized users.

Carbon dioxide versus air insufflation enteroscopy: a systematic review and meta-analysis based on randomized controlled trials

OBJECTIVES

 To compare the insufflation of CO ₂ and ambient air in enteroscopy.

SEARCH SOURCES

 The investigators researched the electronic databases MedLine, Cochrane Library, Central, LILACS, BVS, Scopus and Cinahl. The grey search was conducted in the base of theses of the University of São Paulo, books of digestive endoscopy and references of selected articles and in previous systematic revisions.

STUDY ELIGIBILITY CRITERIA

 The evaluation of eligibility was performed independently, in a non-blind manner, by two reviewers, firstly by title and abstract, followed by complete text. Disagreements between the reviewers were resolved by consensus.

DATA COLLECTION AND ANALYSIS METHOD

 Through the spreadsheet of data extraction, where one author extracted the data and a second author checked the extraction. Disagreements were resolved by debate between the two reviewers. The quality analysis of the studies was performed using the Jadad score. The software RevMan 5 version 5.3 was used for the meta-analysis.

RESULTS

 Four randomized clinical trials were identified, totaling 473 patients submitted to enteroscopy and comparing insufflation of CO ₂ and ambient air. There was no statistical difference in the intubation depth between the two groups. When CO ₂ insufflation was reduced, there was a significant difference in pain levels 1 hour after the procedure (95 % IC, -2.49 [-4.72, -0.26], P : 0.03, I ² : 20%) and 3 hours after the procedure (95% IC, -3.05 [-5.92, -0.18], P : 0.04, I ² : 0 %). There was a usage of lower propofol dosage in the CO ₂ insufflation group, with significant difference (95 % IC, -67.68 [-115.53, -19.84], P : 0.006, I ² : 0 %). There was no significant difference between the groups in relation to the use of pethidine and to the oxygen saturation.

LIMITATIONS

 Restricted number of randomized clinical trials and nonuniformity of data were limitations to the analysis of the outcomes.

CONCLUSION

 The use of CO ₂ as insufflation gas in enteroscopy reduces the pain levels 1 hour and 3 hours after the procedure, in addition to the reduction of the sedation (propofol) dosage used.

Population estimation using the multiple system estimator in the presence of continuous covariates

In the presence of continuous covariates, standard capture-recapture methods assume either that the registrations operate independently at the individual level or that the covariates can be stratified and log-linear models fitted, permitting the modelling of dependence between data sources. This article introduces an approach where direct dependence between registrations is modelled leaving the continuous covariates in their measurement scale. Simulations show that not accounting for possible dependence between registrations results in biased estimation of both the population size and standard error. The proposed method is applied to Dutch neural tube defect registration data.

Capture-Recapture Estimators in Epidemiology with Applications to Pertussis and Pneumococcal Invasive Disease Surveillance

Introduction

Surveillance networks are often not exhaustive nor completely complementary. In such situations, capture-recapture methods can be used for incidence estimation. The choice of estimator and their robustness with respect to the homogeneity and independence assumptions are however not well documented.

Methods

We investigated the performance of five different capture-recapture estimators in a simulation study. Eight different scenarios were used to detect and combine case-information. The scenarios increasingly violated assumptions of independence of samples and homogeneity of detection probabilities. Belgian datasets on invasive pneumococcal disease (IPD) and pertussis provided motivating examples.

Results

No estimator was unbiased in all scenarios. Performance of the parametric estimators depended on how much of the dependency and heterogeneity were correctly modelled. Model building was limited by parameter estimability, availability of additional information (e.g. covariates) and the possibilities inherent to the method. In the most complex scenario, methods that allowed for detection probabilities conditional on previous detections estimated the total population size within a 20–30% error-range. Parametric estimators remained stable if individual data sources lost up to 50% of their data. The investigated non-parametric methods were more susceptible to data loss and their performance was linked to the dependence between samples; overestimating in scenarios with little dependence, underestimating in others. Issues with parameter estimability made it impossible to model all suggested relations between samples for the IPD and pertussis datasets. For IPD, the estimates for the Belgian incidence for cases aged 50 years and older ranged from 44 to58/100,000 in 2010. The estimates for pertussis (all ages, Belgium, 2014) ranged from 24.2 to30.8/100,000.

Conclusion

We encourage the use of capture-recapture methods, but epidemiologists should preferably include datasets for which the underlying dependency structure is not too complex, a priori investigate this structure, compensate for it within the model and interpret the results with the remaining unmodelled heterogeneity in mind.

Estimating the Esophagus Cancer Incidence Rate in Ardabil, Iran: A Capture-Recapture Method

Background:

Accurate cancer registry and awareness of cancer incidence rate is essential in order to define strategies for cancer prevention and control programs. Capture-recapture methods have been recommended for reducing bias and increase the accuracy of cancer incidence estimation.

Objectives:

This study aimed to estimate the esophagus cancer incidence by capture-recapture method based on Ardabil population-based cancer registry data.

Patients and Methods:

Total new cases of esophagus cancer reported by three sources of pathology reports, medical records, and death certificates to Ardabil province cancer registry center in 2006 and 2008 were enrolled in the study. All duplicated cases between three sources were identified and removed using Excel software. Some characteristics such as name, surname, father’s name, date of birth and ICD codes related to their cancer type were used for data linkage and finding the common cases among three sources. The incidence rate per 100,000 was estimated based on capture-recapture method using the log-linear models. We used BIC, G² and AIC statistics to select the best-fit model.

Results:

After removing duplicates, total 471 new cases of esophagus cancer were reported from three sources. The model with linkage between pathology reports, medical record sources and independence with the death certificates source was the best fitted model. The reported incidence rate for the years 2006 and 2008 was 18.77 and 18.51 per 100,000, respectively. In log-linear analysis, the estimated incidence rate for the years 2006 and 2008 was 49.71 and 53.87 per 100,000 populations, respectively.

Conclusions:

Based on the obtained results, it can be concluded that none of the sources of pathology reports, death certificates and medical records individually or collectively were fully covered the incidence cases of esophagus cancer and need to apply some changes in data abstracting and case finding.

Bayesian population size estimation using Dirichlet process mixtures

We introduce a new Bayesian nonparametric method for estimating the size of a closed population from multiple-recapture data. Our method, based on Dirichlet process mixtures, can accommodate complex patterns of heterogeneity of capture, and can transparently modulate its complexity without a separate model selection step. Additionally, it can handle the massively sparse contingency tables generated by large number of recaptures with moderate sample sizes. We develop an efficient and scalable MCMC algorithm for estimation. We apply our method to simulated data, and to two examples from the literature of estimation of casualties in armed conflicts.

Problem drug use prevalence estimation revisited: heterogeneity in capture-recapture and the role of external evidence

BACKGROUND AND AIMS

Capture-recapture (CRC) analysis is recommended for estimating the prevalence of problem drug use or people who inject drugs (PWID). We aim to demonstrate how naive application of CRC can lead to highly misleading results, and to suggest how the problems might be overcome.

METHODS

We present a case study of estimating the prevalence of PWID in Bristol, UK, applying CRC to lists in contact with three services. We assess: (i) sensitivity of results to different versions of the dominant (treatment) list: specifically, to inclusion of non-incident cases and of those who were referred directly from one of the other services; (ii) the impact of accounting for a novel covariate, housing instability; and (iii) consistency of CRC estimates with drug-related mortality data. We then incorporate formally the drug-related mortality data and lower bounds for prevalence alongside the CRC into a single coherent model.

RESULTS

Five of 11 models fitted the full data equally well but generated widely varying prevalence estimates, from 2740 [95% confidence interval (CI) = 2670, 2840] to 6890 (95% CI = 3740, 17680). Results were highly sensitive to inclusion of non-incident cases, demonstrating the presence of considerable heterogeneity, and were sensitive to a lesser extent to inclusion of direct referrals. A reduced data set including only incident cases and excluding referrals could be fitted by simpler models, and led to much greater consistency in estimates. Accounting for housing stability improved model fit considerably more than did the standard covariates of age and gender. External data provided validation of results and aided model selection, generating a final estimate of the number of PWID in Bristol in 2011 of 2770 [95% credible interval (Cr-I) = 2570, 3110] or 0.9% (95% Cr-I = 0.9, 1.0%) of the population aged 15-64 years.

CONCLUSIONS

Steps can be taken to reduce bias in capture-recapture analysis, including: careful consideration of data sources, reduction of lists to less heterogeneous subsamples, use of covariates and formal incorporation of external data.

Estimating the completeness of gastric cancer registration in Ardabil/Iran by a capture-recapture method using population-based cancer registry data

BACKGROUND

Knowledge of cancer incidences is essential for cancer prevention and control programs. Capture-recapture methods have been recommended for reducing bias and increasing the accuracy of cancer incidence estimations. This study aimed to estimate the completeness of gastric cancer registration by the capture-recapture method based on Ardabil population-based cancer registry data.

MATERIALS AND METHODS

All new cases of gastric cancer reported by three sources, pathology reports, death certificates and medical records that reported to Ardabil population-based cancer registry in 2006 and 2008 were enrolled in the study. The duplicate cases based on the similarity of first name, surname and fathers names were identified between sources. The estimated number of gastric cancers was calculated by the log-linear method using Stata 12 software.

RESULTS

A total of 857 new cases of gastric cancer were reported from three sources. After removing duplicates, the reported incidence rates for the years 2006 and 2008 were 35.3 and 32.5 per 100,000 population, respectively. The estimated completeness calculated by log-linear method for these years was 36.7 and 36.0, respectively.

CONCLUSIONS

These results indicate that none of the sources of pathology reports, death certificates and medical records individually or collectively fully cover the incident cases of gastric cancer. We can obtain more accurate estimates of incidence rates using the capture-recapture method.

Incidence of multiple sclerosis in Northern Lisbon, Portugal: 1998-2007

Background

There are few, recent, well assessed, multiple sclerosis (MS) incidence surveys on European populations. This study sought to measure MS incidence in a Northern Lisbon population and assess it using capture-recapture methods (CRMs).

Methods

Among the population residing in the Northern Lisbon Health Area, registered MS diagnoses were obtained from general practitioners in three primary-care districts covering a population of 196,300, and a neurology unit at the main referral hospital. Cases with onset during the periods 1978–1997 and 2008–2012 were excluded due to perceived poor access to image-supported neurological diagnosis and administrative changes in patient referral respectively. Age- and sex-specific incidences for the period 1998–2007 were calculated using McDonald diagnostic criteria, and CRMs were used to correct age-specific incidence rates. The corrected figures were also adjusted for age using the European Standard Population as reference.

Results

When applied to 62 MS patients with onset in the period 1998–2007, the rates per 100,000 population were as follows for both sexes: crude, 3.16; age-adjusted, 3.09 (95% CI 2.32 to 3.87); CRM-adjusted, 4.53 (95% CI 3.13 to 5.94); and age- and CRM-adjusted, 4.48 (3.54-5.41). In general, the rates were 3-fold higher among women than among men. Negative source dependency and CRM impact were highest at ages 35–44 years, where a 60% rise led to a peak incidence.

Conclusions

MS incidence in Northern Lisbon, Portugal, is moderately lower than that yielded by surveys on European populations. CRMs, which in this instance suggest undercounts, are a potentially useful tool for case-finding assessment but their application may introduce bias.

Estimating the size of the population of persons who inject drugs in the island of Montréal, Canada, using a six-source capture-recapture model

BACKGROUND

To plan and implement services to persons who inject drugs (PWID), knowing their number is essential. For the island of Montréal, Canada, the only estimate, of 11,700 PWID, was obtained in 1996 through a capture-recapture method. Thirteen years later, this study was undertaken to produce a new estimate.

METHODS

PWID were defined as individuals aged 14-65 years, having injected recently and living on the island of Montréal. The study period was 07/01/2009 to 06/30/2010. An estimate was produced using a six-source capture-recapture log-linear regression method. The data sources were two epidemiological studies and four drug dependence treatment centres. Model selection was conducted in two steps, the first focusing on interactions between sources and the second, on age group and gender as covariates and as modulators of interactions.

RESULTS

A total of 1480 PWID were identified in the six capture sources. They corresponded to 1132 different individuals. Based on the best-fitting model, which included age group and sex as covariates and six two-source interactions (some modulated by age), the estimated population was 3910 PWID (95% confidence intervals (CI): 3180-4900) which represents a prevalence of 2.8 (95% CI: 2.3-3.5) PWID per 1000 persons aged 14-65 years.

CONCLUSIONS

The 2009-2010 estimate represents a two-third reduction compared to the one for 1996. The multisource capture-recapture method is useful to produce estimates of the size of the PWID population. It is of particular interest when conducted at regular intervals thus allowing for close monitoring of the injection phenomenon.

Relationship between death and infections among patients hospitalized in internal medicine departments: a prevalence and validation study

BACKGROUND

We assessed the burden of infection-related death at internal medicine departments in Denmark and the validity of 2 population-based registries for diagnoses of infection.

METHODS

We reviewed medical records of all inpatients who died at a large university hospital during 2008 with an infection diagnosis in the Cause of Death Registry (CDR) or Hospital Discharge Registry (HDR). We computed the positive predictive value of infection diagnoses and completeness of each registry with 95% confidence intervals (CIs) and the prevalence of infection-related deaths by capture-recapture analysis.

RESULTS

Among 458 deaths, 193 patients (42.1%) had an infection diagnosis of which 40.0% (183 out of 458) were verified. The positive predictive value of an infection diagnosis was 96.0% (95% CI, 92.0%-98.3%) in the CDR and 95.3% (CI, 90.9%-97.9%) in the HDR. Completeness of the CDR was 79.2% (CI, 72.9%-84.6%) and completeness of the HDR was 77.0% (CI, 70.6%-82.7%). By capture-recapture analysis we estimated that 43.2% of all deaths (198 out of 458) were related to infection.

CONCLUSIONS

The large proportion of deaths in internal medicine departments that are related to infection emphasizes the need for optimized infection prevention and treatment strategies. CDR and HDR are valid sources for identifying infection and may supplement each other for increased completeness of infection-related death in epidemiologic research.

Epidemiology of primary Sjögren's syndrome in a French multiracial/multiethnic area

OBJECTIVE

To describe the epidemiology of primary Sjögren's syndrome (SS) in a multiracial/multiethnic population.

METHODS

A cross-sectional study with 5 case-retrieval sources identified adults with primary SS living in the Greater Paris area (population 1,172,482 adults) in 2007. Diagnoses were verified by the American-European Consensus Group (AECG) criteria and study-specific enlarged criteria based on the presence of ≥3 of 4 AECG items among subjective oral or ocular dryness, anti-SSA/SSB positivity, and positive minor salivary gland biopsy results. Prevalence estimates were standardized to those for the world population and a 5-source capture-recapture analysis (CRA) was used. Racial/ethnic differences in primary SS features were evaluated.

RESULTS

In all, 133 subjects met the AECG criteria and 203 met the enlarged criteria. The 2007 prevalence of primary SS was 1.02 cases per 10,000 adults (95% confidence interval [95% CI] 0.85-1.22) for the AECG criteria and 1.52 cases per 10,000 adults (95% CI 1.30-1.76) for the enlarged criteria. The CRA indicated completeness of case findings of ∼90%. Compared to subjects with European backgrounds, those with non-European backgrounds had 2.1-2.3 times higher primary SS prevalence and were younger (P < 0.0001) and were more likely to have polyclonal hypergammaglobulinemia (P < 0.0001) and anti-SSA/SSB antibodies (P = 0.0005 and P < 0.0001 for the AECG and enlarged criteria, respectively).

CONCLUSION

The figure of 1.02–1.52 cases per 10,000 adults we found and estimates from the few other population-based census surveys support that the prevalence of diagnosed primary SS is between 1 and 9 cases per 10,000 (0.01-0.09%) [corrected] in the general population. Non-European race/ethnicity may be associated with increased primary SS risk and a distinct disease profile.

The hidden population: Some methodological issues about estimation of problematic drug use

Aim

This article analyses and discusses the estimation of serious or problematic drug use through an empirical example based on a local Swedish study in Gothenburg.

Methods and Data

This was a case-finding study with questionnaires directed at organisations which have contact with the target group. The material was supplemented with information from the two documentation systems DOK and ASI. A total of 2,148 reports were collected. Health care data of 1,096 individuals was also collected for analysis with the truncated Poisson method. Analyses with capture-recapture or truncated Poisson were conducted to calculate the size of the hidden population.

Results

The statistical analyses resulted in variable numbers for the hidden population, and the total prevalence of serious drug abuse in Gothenburg is estimated to be between 2,200 and 4,400 people.

Conclusion

The study shows that estimation of the presence and prevalence of problematic drug abuse involves many methodological difficulties and challenges. The significant variation of the size of the hidden population presented in the study raises doubts about the reliability and validity of the different methods. The methods are clearly sensitive to the importance of fulfilling the different basic assumptions.

Quality assessment of capture-recapture studies in resource-limited countries

OBJECTIVES

Resource-limited countries often lack robust routine surveillance systems to accurately assess the burden of human attributes and diseases. In these settings capture-recapture analysis can be an alternative tool to obtain prevalence and incidence rates. Performance of capture-recapture analyses in resource-limited countries has not been systematically reviewed.

METHODS

Systematic review of the performance of capture-recapture analyses in the categories of human attributes, non-infectious and infectious diseases in resource-limited countries, assessing individual study quality criteria and a minimum quality criterion per category, using PRISMA methodology.

RESULTS

A total of 1671 potentially relevant PubMed citations were screened, resulting in 52 eligible publications: 36% in human attributes, i.e. hidden populations, injuries and mortality; 48% in non-infectious and 15% in infectious disease categories. Twenty-one per cent of selected studies were from low income countries, 40% from lower-middle-income countries and 38% from upper-middle-income countries. Thirteen per cent achieved good individual study quality criteria, 25% were intermediate and 19% were poor. Of the good studies, six were performed on human attributes and one on a non-infectious disease. The proportions of publications meeting the minimum quality criterion per category were 42%, 20% and 37%, respectively.

CONCLUSIONS

Few capture-recapture studies in resource-limited countries achieved good individual quality criteria and a minority met the minimum quality criterion per category. Capture-recapture techniques in these settings should be carefully considered and implemented rigorously and are not a panacea for strengthening of routine surveillance systems.

Using capture-recapture methods to better ascertain the incidence of fatal child maltreatment

OBJECTIVES

To (1) test the use of capture-recapture methods to estimate the total number of child maltreatment deaths in a single state using information from death certificates, child welfare reports, child death review teams, and uniform crime reports; and to (2) compare these estimates to the number of maltreatment deaths identified through an in-depth "gold standard" review.

METHODS

Child maltreatment deaths were identified in four existing administrative data sources: (1) death reports in our state vital statistics (DC); (2) child death review team reports (CDR); (3) homicide reports filed by our state police agency as uniform crime report (UCR) supplements for the FBI; and (4) abstracted reports of a minor's death from our state child protective services (CPS) agency. Capture-recapture pair-wise and pooled comparisons were then applied to estimate the numbers of abuse and total maltreatment deaths and were compared to the number of cases identified by independent case review.

RESULTS

There were a total of 194 child maltreatment deaths in Michigan during 2000-2001 with 66 due to physical abuse. Capture-recapture analysis estimated the mean number of total child maltreatment deaths as 101.02 (95%CI=92.52, 109.53), with abuse deaths of 64.55 (60.85, 68.25). Most pair-wise and pooled comparisons worked equally well for abuse deaths, but estimates for total child maltreatment deaths were low.

CONCLUSIONS

Capture-recapture methods applied to existing administrative datasets produced accurate estimates of child abuse deaths but were not useful in producing reliable estimates of total child maltreatment deaths due to undercounting neglect-related deaths in all existing administrative data sets. The underlying assumptions for capture-recapture methods were not met for neglect deaths. Local and/or state teams conducting ongoing intensive case review may yet remain the best way to identify the total number of child maltreatment deaths.

PRACTICE IMPLICATIONS

Capture-recapture methods allow for more accurate estimation of the true number of child physical abuse deaths than does using single existing sources of child fatality information, but deaths from causes other than abuse are undercounted. Child maltreatment fatality surveillance requires a systematic process and standard criteria for identifying cases of maltreatment, particularly neglect-related child deaths.

Estimating the prevalence of schizophrenia among New Zealand Maori: a capture-recapture approach

OBJECTIVE

The aim of the present study was to estimate the 12 month prevalence of schizophrenia in the Maori population of New Zealand.

METHOD

Mental health data from two national sources were obtained for the period 2000-2003. A simple count of unique individuals with schizophrenia was used to estimate contact prevalence and a four-list capture-recapture procedure to estimate population prevalence.

RESULTS

Contact prevalence was significantly lower than the estimated population prevalence for both groups. The estimated 12 month prevalence of schizophrenia for Maori (0.97%) was significantly higher than for non-Maori (0.32%), even after adjustment for age, case under-ascertainment, and socioeconomic deprivation.

CONCLUSIONS

The prevalence of schizophrenia among Maori appears to be elevated, although limitations in diagnostic reliability and recording of ethnicity must be considered. This adds further evidence of worldwide variation in the prevalence of schizophrenia. Capture-recapture provides a reliable cost-effective alternative to epidemiological surveys for estimating the prevalence of low-prevalence disorders such as schizophrenia.

Incidence and completeness of notification of Legionnaires' disease in The Netherlands: covariate capture-recapture analysis acknowledging regional differences

To estimate incidence and completeness of notification of Legionnaires' disease (LD) in The Netherlands in 2000 and 2001, we performed a capture-recapture analysis using three registers: Notifications, Laboratory results and Hospital admissions. After record-linkage, 373 of the 780 LD patients identified were notified. Ascertained under-notification was 52.2%. Because of expected and observed regional differences in the incidence rate of LD, alternatively to conventional log-linear capture-recapture models, a covariate (region) capture-recapture model, not previously used for estimating infectious disease incidence, was specified and estimated 886 LD patients (95% confidence interval 827-1022). Estimated under-notification was 57.9%. Notified, ascertained and estimated average annual incidence rates of LD were 1.15, 2.42 and 2.77/100 000 inhabitants respectively, with the highest incidence in the southern region of The Netherlands. Covariate capture-recapture analysis acknowledging regional differences of LD incidence appears to reduce bias in the estimated national incidence rate.

Record-linkage and capture-recapture analysis to estimate the incidence and completeness of reporting of tuberculosis in England 1999-2002

In 1999 the Enhanced Tuberculosis Surveillance (ETS) system was introduced in the United Kingdom to strengthen surveillance of tuberculosis (TB). The aim of this study was to assess the use of record-linkage and capture-recapture methodology for estimating the completeness of TB reporting in England between 1999 and 2002. Due to the size of the TB data sources sophisticated record-linkage software was required and the proportion of false-positive cases among unlinked hospital-derived TB records was estimated through a population mixture model. This study showed that record-linkage of TB data sources and cross-validation with additional TB-related datasets improved data quality as well as case ascertainment. Since the introduction of ETS observed completeness of notification in England has increased and the results were consistent with expected levels of under-notification. Completeness of notification estimated by a log-linear capture-recapture model was highly inconsistent with prior estimates and the validity of this methodology was further examined.

Completeness of notification of tuberculosis in The Netherlands: how reliable is record-linkage and capture-recapture analysis?

The aim of this study was to describe a systematic process of record-linkage, cross-validation, case-ascertainment and capture-recapture analysis to assess the quality of tuberculosis registers and to estimate the completeness of notification of incident tuberculosis cases in The Netherlands in 1998. After record-linkage and cross-validation 1499 tuberculosis patients were identified, of whom 1298 were notified, resulting in an observed under-notification of 13.4%. After adjustment for possible imperfect record-linkage and remaining false-positive hospital cases observed under-notification was 7.3%. Log-linear capture-recapture analysis initially estimated a total number of 2053 (95% CI 1871-2443) tuberculosis cases, resulting in an estimated under-notification of 36.8%. After adjustment for possible imperfect record-linkage and remaining false-positive hospital cases various capture-recapture models estimated under-notification at 13.6%. One of the reasons for the higher than expected estimated under-notification in a country with a well-organized system of tuberculosis control might be that some tuberculosis cases, e.g. extrapulmonary tuberculosis, are managed by clinicians less familiar with notification of infectious diseases. This study demonstrates the possible impact of violation of assumptions underlying capture-recapture analysis, especially the perfect record-linkage, perfect positive predictive value and absent three-way interaction assumptions.

A conditional probability approach to surveillance system sensitivity assessment

OBJECTIVE

To determine the sexually transmitted diseases (STD) surveillance system sensitivity with a conditional probability approach at district level in Darregaz, a frontier town in the north of Iran.

STUDY DESIGN

A cross-sectional survey.

METHODS

We used a sample survey of sexually active inhabitants for proxy measurement of the medical service utilization pattern for STD, and interviews with all practitioners to determine their knowledge of STD diagnosis and attitude towards STD reporting as proxy measures of actual STD diagnosis and reporting, respectively. Point estimates of the STD surveillance system sensitivity for each of the health service sectors were derived from multiplying the three proxy measures of sensitivity determinants, i.e., utilization, diagnosis, and reporting, as conditional probabilities. Estimates of sensitivity for all health service sectors were summed to obtain the overall sensitivity.

RESULTS

The sensitivity of the surveillance system was 21.2% (95% confidence interval (CI) 15.5-25.3%) for detecting symptomatic STD. Of the sexually active inhabitants, 8.9% (95% CI 5.5-14.2%) did not use health services if they contracted STDs. The public health sector's contribution to overall sensitivity (59.6%) was greater than its proportion of service utilization for STD (45.3%).

CONCLUSIONS

The strengths of the conditional probability approach are feasibility of conducting necessary surveys, decomposing sensitivity into its determinants, and providing evidence for intervention at different points for planning purposes. This approach tends to overestimate the overall sensitivity.

Estimating infectious diseases incidence: validity of capture-recapture analysis and truncated models for incomplete count data

Capture-recapture analysis has been used to evaluate infectious disease surveillance. Violation of the underlying assumptions can jeopardize the validity of the capture-recapture estimates and a tool is needed for cross-validation. We re-examined 19 datasets of log-linear model capture-recapture studies on infectious disease incidence using three truncated models for incomplete count data as alternative population estimators. The truncated models yield comparable estimates to independent log-linear capture-recapture models and to parsimonious log-linear models when the number of patients is limited, or the ratio between patients registered once and twice is between 0.5 and 1.5. Compared to saturated log-linear models the truncated models produce considerably lower and often more plausible estimates. We conclude that for estimating infectious disease incidence independent and parsimonious three-source log-linear capture-recapture models are preferable but truncated models can be used as a heuristic tool to identify possible failure in log-linear models, especially when saturated log-linear models are selected.

Underreporting of Meningococcal Disease Incidence in the Netherlands: Results from a Capture–Recapture Analysis Based on Three Registration Sources with Correction for False Positive Diagnoses

In order to come to a reliable evaluation of the effectiveness of the chosen vaccination policy regarding meningococcal disease, the completeness of registrations on meningococcal disease in the Netherlands was estimated with the capture-recapture method. Data over 1993-1998 were collected from (A) mandatory notifications (n = 2926); (B) hospital registration (n = 3968); (C) laboratory surveillance (n = 3484). As the standard capture-recapture method does not take into account false positive diagnoses, we developed a model to adjust for the lack of specificity of our sources. We estimated that 1363 cases were not registered in any of the three sources in the period of study. The completeness of the three sources was therefore estimated at 49% for source A, 67% for source B and 58% for source C. After adjustment for false positive diagnoses, the completeness of source A, B, and C was estimated as 52%, 70% and 62%, respectively. The capture-recapture methods offer an attractive approach to estimate the completeness of surveillance sources and hence contribute to a more accurate estimate of the disease burden under study. However, the method does not account for higher-order interactions or presence of false positive diagnoses. Being aware of these limitations, the capture-recapture method still elucidates the (in)completeness of sources and gives a rough estimate of this (in)completeness. This makes a more accurate monitoring of disease incidence possible and hence attributes to a more reliable foundation for the design and evaluation of health interventions such as vaccination programs.

Capturing crack cocaine use: estimating the prevalence of crack cocaine use in London using capture-recapture with covariates

AIM

To estimate the prevalence of crack cocaine use in 12 London Boroughs (and London as a whole).

SETTING

Twelve London Boroughs, 2000-01.

METHODS

(1) Covariate capture-recapture techniques applied to three data sources of subjects reporting crack cocaine use: specialist drug treatment (2905), arrest referral (1188) and accident and emergency and community survey (531); and (2) ratio-estimation multiplier, using an estimate of number of injecting drug users and proportion that use crack cocaine.

FINDINGS

After matching, 4117 individuals aged 15-44 were identified. The best-fitting model estimated 16 855 unobserved crack cocaine users, giving an overall estimate of approximately 21 000 [95% confidence interval (CI) 13 000-43,000] and a prevalence of 1.5% (95% CI 1.0-3.2%). Prevalence of crack cocaine use was 2.4% (95% CI 1.5-5.0%) among men and 0.7% (95% CI 0.5-1.0%) among women, and similar by age groups 15-29 and 30-44 years. Overall, approximately 11 900 (57%) of the estimated number of crack cocaine users were also opiate users. In London as a whole there may be 46,000 (1.3%) crack cocaine users aged 15-44 years, with 28 000 (1.9%) in inner London-four times higher than estimates from population surveys. Some corroboration was provided by the ratio-estimation method, which estimated 23 000 users in the 12 Boroughs.

CONCLUSIONS

Capture-recapture can be applied to crack cocaine and obtain better estimates than population surveys. The size of the crack cocaine-using population in London is large, although currently the majority are also opiate users. Given that half of current users are under 30 the problems associated with crack cocaine use are likely to increase in the future.

Prior information in behavioral capture-recapture methods: demographic influences on drug injectors' propensity to be listed in data sources and their drug-related mortality

The authors present findings from a Bayesian analysis of Scotland's four primary capture-recapture data sources for 2000 that was carried out to estimate numbers of current injecting drug users by region (Greater Glasgow vs. elsewhere in Scotland), sex (male vs. female), and age group (15-34 years vs. > or =35 years). A secondary goal of the analysis was to obtain Bayesian estimates and credible intervals for the demographic influences on Scotland's drug-related death rate per 100 current injectors. Incorporation of informative priors altered the models with highest posterior probability. Expert opinion on how demography influenced Scottish drug injectors' propensity to be listed in different data sources was taken into account, along with external information about European injectors' drug-related death rates and male:female ratios. Higher drug-related mortality was confirmed in older drug injectors and those outside of Greater Glasgow. Female injectors' lower drug-related death rate was not sustained beyond 34 years of age. The authors recommend that demographic influences be accommodated in behavioral capture-recapture estimation, especially when it is a prelude to secondary analysis, such as the analysis of drug-related death rates presented here.

Injecting drug use in Brighton, Liverpool, and London: best estimates of prevalence and coverage of public health indicators

STUDY OBJECTIVE

To estimate the prevalence of injecting drug use (IDU) in three cities in England and to measure the coverage of key public health indicators.

DESIGN

Capture-recapture techniques with covariate effects.

SETTING

Liverpool, Brighton, and 12 London boroughs, 2000/01.

PARTICIPANTS

IDU collated and matched across five data sources-community recruited survey, specialist drug treatment, arrest referral, syringe exchange, and accident and emergency-896 in Brighton, 1224 in Liverpool, and 6111 in London.

MAIN RESULTS

It is estimated that in 2000/01 the number and prevalence of IDU aged 15-44 was 2300 (95%CI 1500 to 3700) and 2.0% (95%CI% 1.3% to 3.2%) in Brighton; 2900 (95%CI 2500 to 5000) and 1.5% (95%CI 1.3% to 2.6%) in Liverpool; 16 700 (95%CI 13 800 to 21 600) and 1.2% (95%CI 1.0% to 1.6%) in 12 London boroughs; with a prevalence of 1.7% (95%CI 1.2% to 3.3%) in inner London. It is estimated that: less than one in four IDU are in treatment in the three areas; syringe exchange programmes covered about 25% of injections in Brighton and Liverpool and 20% in London; and that the annual opioid mortality rate among IDU was 2% in Brighton compared with less than 1% in Liverpool and London.

CONCLUSIONS

Credible estimates of the prevalence of injecting drug use (and key public health indicators) can be determined using covariate capture-recapture techniques. These suggest that: targets to double the number in treatment are possible: syringe distribution should be increased; and further attention, especially in Brighton, given to reducing overdose mortality.

The multiple-record systems estimator when registrations refer to different but overlapping populations

In multiple-record systems estimation it is usually assumed that all registration relate to the same population. In this paper, we develop a method which can be used when the registrations relate to different populations, in the sense that they cover, for example, different time periods or regions. We show that under certain conditions ignoring that the registrations relate to different populations results in correct estimates of population size. The EM algorithm is presented as a method that can be used for more general problems. The parametric bootstrap is used to construct a confidence interval. The proposed method is then applied to a data set with five registrations of neural tube defects, that cover different time periods.

Recent transmission of tuberculosis in Madrid: application of capture–recapture analysis to conventional and molecular epidemiology

BACKGROUND

Population-based studies using a combination of molecular techniques and conventional epidemiological methods have been used to study the dynamics of tuberculosis (TB) transmission but the relative utility of each technique has not yet been established.

METHODS

A prospective population-based molecular and epidemiological study of patients diagnosed with TB was conducted in three urban districts of Madrid (Spain) during 1997-1999. Analysis was performed using the capture-recapture method including covariates in which conventional epidemiological data and the information on clustered cases obtained by DNA fingerprinting were regarded as independent and complementary procedures.

RESULTS

The estimate obtained by molecular analysis alone, that 31.6% of TB cases were due to recent transmission, was revised to 44.8% (95% CI: 31.4-58.2) using the capture-recapture method. The estimated completeness of the combined databases for identification of recent transmission was 59.2%. Underestimation of the true prevalence of recent transmission was higher with conventional epidemiology than molecular analysis, particularly for patients <35 years old and those with a history of imprisonment.

CONCLUSIONS

In this study, use of the capture-recapture technique allowed us to combine epidemiological information obtained by conventional and molecular methods to quantify the number of cases of recently transmitted TB in the community and identify specific populations at high risk of disease. This information is clearly important because such groups are a prime target for improved TB control measures. In the long term, this combination of techniques may contribute significantly to control the spread of TB.

Estimating the total number of newly-recognized silicosis cases in the United States

BACKGROUND

The US employer-based surveillance system for documenting occupational injuries and illnesses undercounts chronic diseases. We suggest a method to estimate the number of individuals who are newly-recognized with silicosis each year in the United States.

METHODS

Data from US death certificates, the Michigan state-based surveillance system, and capture-recapture analysis were used to calculate national estimates of silicosis.

RESULTS

From 1987 to 1996, 2,787 deaths occurred in the United States where silicosis was mentioned on the death certificates. During the same period, in Michigan 77% of death certificates with a mention of silicosis were confirmed as silicosis-related deaths and the ratio of the number of living to deceased confirmed silicosis cases was 6.44. The proportion of confirmed silicosis deaths, the ratio of the living to deceased silicosis cases and capture-recapture analysis from the Michigan surveillance system, were used to estimate that there were 3,600-7,300 cases per year of silicosis in the United States from 1987 to 1996.

CONCLUSIONS

Our estimate of the annual number of newly-recognized silicosis cases is significantly larger than the estimate from the employer-based reporting system used for counting occupational disease in the United States. This employer-based surveillance system is inadequate for determining the frequency of occupational disease. Our analysis which combines a readily-available and relatively inexpensive national administrative database (i.e., death certificates) with a more costly state-based active surveillance system is a cost-effective model that could be used to provide better estimates of a number of different occupational diseases. Accurate estimates of occupational illnesses are essential to both determine temporal trends and evaluate efforts to prevent silicosis.

Prevalence, incidence, survival, and disease characteristics of systemic sclerosis in a large US population

OBJECTIVE

To estimate the prevalence, incidence, survival, and disease characteristics of systemic sclerosis (SSc) in the Detroit tricounty area.

METHODS

A census of SSc cases for the period 1989-1991 was conducted in the Detroit area, using multiple sources for case identification. Diagnoses were verified by medical record review. Capture-recapture analysis was used to estimate the total SSc population. Cases of localized scleroderma (morphea and linear disease) were excluded.

RESULTS

Based on 706 verified cases of SSc, prevalence was initially estimated to be 242.0 cases per million adults (95% confidence interval [95% CI] 213-274), with an annual incidence of 19.3 new cases per million adults per year (95% CI 12.4-30.2). Capture-recapture analysis, based on the degree of overlap of verified cases among multiple sources, resulted in a revised prevalence estimate of 276 cases per million adults (95% CI 245-310). Sex- and race-specific prevalence estimates were significantly higher for women than for men, and for blacks than for whites. The average age at diagnosis was significantly younger for blacks than for whites. Compared with white patients, black patients were almost twice as likely to have diffuse disease (prevalence proportion ratio 1.86, 95% CI 1.48-2.35). Median survival was approximately 11 years. Factors negatively affecting survival included male sex (hazard ratio 1.81, 95% CI 1.29-2.55) and older age at diagnosis (hazard ratio 1.04, 95% CI 1.03-1.05).

CONCLUSION

This study establishes baseline estimates of SSc occurrence and characteristics in a large US cohort consisting primarily of black adults and white adults. These data should facilitate research regarding the role of geographic, ethnic, racial, and environmental factors for this disease in comparison populations.

The applications of capture-recapture models to epidemiological data

Capture-recapture methodology, originally developed for estimating demographic parameters of animal populations, has been applied to human populations. This tutorial reviews various closed capture-recapture models which are applicable to ascertainment data for estimating the size of a target population based on several incomplete lists of individuals. Most epidemiological approaches merging different lists and eliminating duplicate cases are likely to be biased downwards. That is, the final merged list misses those who are in the population but were not ascertained in any of the lists. If there are no matching errors, then the duplicate information collected from a capture-recapture experiment can be used to estimate the number of missed under proper assumptions. Three approaches and their associated estimation procedures are introduced: ecological models; log-linear models, and the sample coverage approach. Each approach has its unique way of incorporating two types of source dependencies: local (list) dependence and dependence due to heterogeneity. An interactive program, CARE (for capture-recapture) developed by the authors is demonstrated using four real data sets. One set of data deals with infection by the acute hepatitis A virus in an outbreak in Taiwan; the other three sets are ascertainment data on diabetes, spina bifida and infants' congenital anomaly discussed in the literature. These data sets provide examples to show the usefulness of the capture-recapture method in correcting for under-ascertainment. The limitations of the methodology and some cautionary remarks are also discussed.

An application of the two-source capture-recapture method to estimate the completeness of the Tuscany Cancer Registry, Italy

The purpose of this study was to apply a two-way capture-recapture method to estimate the Tuscany Cancer Registry completeness, taking into account the presence of dependence between sources. Cases incident during 1995-1996 were flagged according to three sources of information: clinical notes, pathological reports and death certificates. For each group of cases notified by one source the dependence between the other two has been quantified and the completeness has been estimated by a two-way capture-recapture method. When only two (or substantially two) sources are dependent on each other it is possible to correct for it by pooling the two sources in a single group and comparing it with the remainder source by a two-way capture-recapture method. The capture-recapture method has been applied to the overall 12 387 incident cases and to 1569 female breast and 1443 lung cancer cases. After correction for the greatest dependence among the three couples of sources of information, the estimates of completeness were 97.4% for the whole case series, 88.7% for female breast and 99.6% for lung cancer. With the limit of multiple strong dependence between sources, the two-way capture-recapture method seems a simple and useful tool for estimating the completeness of cancer registration.