Dataset Information

When does attrition lead to biased estimates of alcohol consumption? Bias analysis for loss to follow-up in 30 longitudinal cohorts.

ABSTRACT: Objectives: Survey nonresponse has increased across decades, making the amount of attrition a focus in generating inferences from longitudinal data. Use of inverse probability weights [IPWs] and other statistical approaches are common, but residual bias remains a threat. Quantitative bias analysis for nonrandom attrition as an adjunct to IPW may yield more robust inference.

Methods: Data were drawn from the Monitoring the Future panel studies [twelfth grade, base-year: 1976-2005; age 29/30 follow-up: 1987-2017, N = 73,298]. We then applied IPW imputation in increasing percentages, assuming varying risk differences [RDs] among nonresponders. Measurements included past-two-week binge drinking at base-year and every follow-up. Demographic and other correlates of binge drinking contributed to IPW estimation.

Results: Attrition increased: 31.14%, base-year 1976; 61.33%, base-year 2005. The magnitude of bias depended not on attrition rate but on prevalence of binge drinking and RD among nonrespondents. The probable range of binge drinking among nonresponders was 12-45%. In every scenario, base-year and follow-up binge drinking were associated. The likely range of true RDs was 0.14 [95% CI: 0.11-0.17] to 0.28 [95% CI: 0.25-0.31].

Conclusions: When attrition is present, the amount of attrition alone is insufficient to understand contribution to effect estimates. We recommend including bias analysis in longitudinal analyses.

SUBMITTER: Keyes KM

PROVIDER: S-EPMC7723204 | biostudies-literature | 2020 Dec

REPOSITORIES: biostudies-literature

ACCESS DATA

Publications

When does attrition lead to biased estimates of alcohol consumption? Bias analysis for loss to follow-up in 30 longitudinal cohorts.

Keyes Katherine M KM Jager Justin J Platt Jonathan J Rutherford Caroline C Patrick Megan E ME Kloska Deborah D DD Schulenberg John J

International journal of methods in psychiatric research 20200713 4

<h4>Objectives</h4>Survey nonresponse has increased across decades, making the amount of attrition a focus in generating inferences from longitudinal data. Use of inverse probability weights [IPWs] and other statistical approaches are common, but residual bias remains a threat. Quantitative bias analysis for nonrandom attrition as an adjunct to IPW may yield more robust inference.<h4>Methods</h4>Data were drawn from the Monitoring the Future panel studies [twelfth grade, base-year: 1976-2005; ag ...[more]

PMID: 32656917

Similar Datasets

Project description:Attrition is a potential source of bias in cohort studies. Although attrition may be inevitable in cohort studies of older people, there is little empirical evidence as to whether bias due to such attrition is also inevitable. Anonymised primary care data, routinely collected in clinical practice and independent of any cohort research study, represents an ideal unselected comparison dataset with which to compare primary care data from consenting responders to a cohort study. Our objective was to use this method as a novel means to assess if (i) responders at follow-up stages in a cohort study remain representative of responders at baseline and (ii) attrition biases estimates of longitudinal associations. We compared primary care consultation morbidities and prescription prevalences among circa 32,000 patients aged 50+ who contribute to an anonymised general practice database (Consultations in Primary Care Archive (CiPCA)) with those from patients aged 50+ in the North Staffordshire Osteoarthritis Project (NorStOP) cohort, United Kingdom (2002-2008; n=16,159). 8,197 (51%) persons responded to the NorStOP baseline survey and consented to medical record review. 5,121 and 3,311 responded at 3- and 6-year follow-ups. Differences in consulting prevalence of non-musculoskeletal morbidities between NorStOP responders and CiPCA comparison population did not increase over the two follow-up points except for ischaemic heart disease. Differences observed at baseline for osteoarthritis-related consultations were generally unchanged at the two follow-ups (standardised prevalence ratios for osteoarthritis (1.09-1.13) and joint pain (1.12-1.23)). Age and gender adjusted associations between baseline consultation for chronic morbidity and future new osteoarthritis and related consultations were similar in CiPCA (adjusted Hazard Ratio: 1.40; 95% Confidence Interval: 1.34,1.47) and NorStOP 6-year responders (1.32; 1.15,1.51). There was little evidence that responders at follow-ups represented any further selection bias to that present at baseline. Attrition in cohort studies of older people does not inevitably indicate bias.

Project description:PurposeTo examine the association of alcohol consumption and type of alcoholic beverage with incident cataract surgery in 2 large cohorts.DesignLongitudinal, observational study.ParticipantsWe included 469 387 participants of UK Biobank with a mean age of 56 years and 23 162 participants of European Prospective Investigation of Cancer (EPIC)-Norfolk with a mean age of 59 years.MethodsSelf-reported alcohol consumption at baseline was ascertained by a touchscreen questionnaire in UK Biobank and a food-frequency questionnaire in EPIC-Norfolk. Cases were defined as participants undergoing cataract surgery in either eye as ascertained via data linkage to National Health Service procedure statistics. We excluded participants with cataract surgery up to 1 year after the baseline assessment visit or those with self-reported cataract at baseline. Cox proportional hazards models were used to examine the associations of alcohol consumption with incident cataract surgery, adjusted for age, sex, ethnicity, Townsend deprivation index, body mass index (BMI), smoking, and diabetes status.Main outcome measuresIncident cataract surgery.ResultsThere were 19 011 (mean cohort follow-up of 95 months) and 4573 (mean cohort follow-up of 193 months) incident cases of cataract surgery in UK Biobank and EPIC-Norfolk, respectively. Compared with nondrinkers, drinkers were less likely to undergo cataract surgery in UK Biobank (hazard ratio [HR], 0.89; 95% confidence interval [CI], 0.85-0.93) and EPIC-Norfolk (HR, 0.90; 95% CI, 0.84-0.97) after adjusting for covariables. Among alcohol consumers, greater alcohol consumption was associated with a reduced risk of undergoing cataract surgery in EPIC-Norfolk (P < 0.001), whereas a U-shaped association was observed in the UK Biobank. Compared with nondrinkers, subgroup analysis by type of alcohol beverage showed the strongest protective association with wine consumption; the risk of incident cataract surgery was 23% and 14% lower among those in the highest category of wine consumption in EPIC-Norfolk and UK Biobank, respectively.ConclusionsOur findings suggest a lower risk of undergoing cataract surgery with low to moderate alcohol consumption. The association was particularly apparent with wine consumption. We cannot exclude the possibility of residual confounding, and further studies are required to determine whether this association is causal in nature.

Project description:Background. Attrition is a major cause of potential bias in longitudinal studies and clinical trials. Attrition rate above 20% raises concern of the reliability of the results. Few studies have looked at the factors behind attrition in follow-ups spanning decades. Methods. We analyzed attrition and associated factors of a 30-year follow-up cohort of subjects who were born with perinatal risks for neurodevelopmental disorders. Attrition rates were calculated at different stages of follow-up and differences between responders and non-responders were tested. To find combinations of variables influencing attrition and investigate their relative importance at birth, 5, 9, 16 and 30 years of follow-up we used the random forest classification. Results. Initial loss of potential participants was 13%. Attrition was 16% at five, 24% at nine, 35% at 16 and 46% at 30 years. The only group difference that emerged between responders and non-responders was in socioeconomic status (SES). The variables identified by random forest classification analysis were classified into Birth related, Development related and SES related. Variables from all these categories contributed to attrition, but SES related variables were less important than birth and development associated variables. Classification accuracy ranged between 0.74 and 0.96 depending on age. Discussion. Lower SES is linked to attrition in many studies. Our results point to the importance of the growth and development related factors in a longitudinal study. Parents' decisions to participate depend on the characteristics of the child. The same association was also seen when the child, now grown up, decided to participate at 30 years. In addition, birth related medical variables are associated with the attrition still at the age of 30. Our results using a data mining approach suggest that attrition in longitudinal studies is influenced by complex interactions of a multitude of variables, which are not necessarily evident using other multivariate techniques.

Project description:BackgroundAn increasingly popular measure for summarising cancer prognosis is the loss in life expectancy (LLE), i.e. the reduction in life expectancy following a cancer diagnosis. The proportion of life lost (PLL) can also be derived, improving comparability across age groups as LLE is highly age-dependent. LLE and PLL are often used to assess the impact of cancer over the remaining lifespan and across groups (e.g. socioeconomic groups). However, in the presence of screening, it is unclear whether part of the differences across population groups could be attributed to lead time bias. Lead time is the extra time added due to early diagnosis, that is, the time from tumour detection through screening to the time that cancer would have been diagnosed symptomatically. It leads to artificially inflated survival estimates even when there are no real survival improvements.MethodsIn this paper, we used a simulation-based approach to assess the impact of lead time due to mammography screening on the estimation of LLE and PLL in breast cancer patients. A natural history model developed in a Swedish setting was used to simulate the growth of breast cancer tumours and age at symptomatic detection. Then, a screening programme similar to current guidelines in Sweden was imposed, with individuals aged 40-74 invited to participate every second year; different scenarios were considered for screening sensitivity and attendance. To isolate the lead time bias of screening, we assumed that screening does not affect the actual time of death. Finally, estimates of LLE and PLL were obtained in the absence and presence of screening, and their difference was used to derive the lead time bias.ResultsThe largest absolute bias for LLE was 0.61 years for a high screening sensitivity scenario and assuming perfect screening attendance. The absolute bias was reduced to 0.46 years when the perfect attendance assumption was relaxed to allow for imperfect attendance across screening visits. Bias was also present for the PLL estimates.ConclusionsThe results of the analysis suggested that lead time bias influences LLE and PLL metrics, thus requiring special consideration when interpreting comparisons across calendar time or population groups.

Project description:BackgroundThe goals of our study are to determine the most appropriate model for alcohol consumption as an exposure for burden of disease, to analyze the effect of the chosen alcohol consumption distribution on the estimation of the alcohol Population- Attributable Fractions (PAFs), and to characterize the chosen alcohol consumption distribution by exploring if there is a global relationship within the distribution.MethodsTo identify the best model, the Log-Normal, Gamma, and Weibull prevalence distributions were examined using data from 41 surveys from Gender, Alcohol and Culture: An International Study (GENACIS) and from the European Comparative Alcohol Study. To assess the effect of these distributions on the estimated alcohol PAFs, we calculated the alcohol PAF for diabetes, breast cancer, and pancreatitis using the three above-named distributions and using the more traditional approach based on categories. The relationship between the mean and the standard deviation from the Gamma distribution was estimated using data from 851 datasets for 66 countries from GENACIS and from the STEPwise approach to Surveillance from the World Health Organization.ResultsThe Log-Normal distribution provided a poor fit for the survey data, with Gamma and Weibull distributions providing better fits. Additionally, our analyses showed that there were no marked differences for the alcohol PAF estimates based on the Gamma or Weibull distributions compared to PAFs based on categorical alcohol consumption estimates. The standard deviation of the alcohol distribution was highly dependent on the mean, with a unit increase in alcohol consumption associated with a unit increase in the mean of 1.258 (95% CI: 1.223 to 1.293) (R2 = 0.9207) for women and 1.171 (95% CI: 1.144 to 1.197) (R2 = 0. 9474) for men.ConclusionsAlthough the Gamma distribution and the Weibull distribution provided similar results, the Gamma distribution is recommended to model alcohol consumption from population surveys due to its fit, flexibility, and the ease with which it can be modified. The results showed that a large degree of variance of the standard deviation of the alcohol consumption Gamma distribution was explained by the mean alcohol consumption, allowing for alcohol consumption to be modeled through a Gamma distribution using only average consumption.

Dataset Information

When does attrition lead to biased estimates of alcohol consumption? Bias analysis for loss to follow-up in 30 longitudinal cohorts.

Publications

When does attrition lead to biased estimates of alcohol consumption? Bias analysis for loss to follow-up in 30 longitudinal cohorts.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets