Using Propensity Score Weighting to Reduce Selection Bias in Large-Scale Data Sets

Bishop, Crystal D.; Leite, Walter L.; Snyder, Patricia

doi:10.1177/1053815118793430

Cited by 26 publications

(16 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…We used IPW to minimise selection bias due to responsiveness to the questionnaire in which maltreatment was reported. 27 This resulted in a good balance of the measured covariates between those who responded and did not respond to the online questionnaire (see online supplementary eFigure 3), therefore, reducing bias due to selective participation. Of note, unweighted estimates were similar to weighted estimates, indicating that bias due to selection to the online questionnaire did not substantially affect results (results available from authors on request).…”

Section: Strengths and Limitationsmentioning

confidence: 99%

Sex differences in the association between childhood maltreatment and cardiovascular disease in the UK Biobank

Soares

Hammerton

Howe

et al. 2020

Heart

View full text Add to dashboard Cite

ObjectivesTo assess and compare associations between childhood maltreatment and cardiovascular disease (CVD) in men and women in the UK. In secondary analyses, we also explored possible age differences and associations with early onset CVD (<50 years).MethodsWe included 157 311 participants from the UK Biobank who had information on physical, sexual or emotional abuse, emotional or physical neglect. CVD outcomes were defined as any CVD, hypertensive disease, ischaemic heart disease (IHD) and cerebrovascular disease. These were extracted from self-report, blood pressure measurements, hospital register and death register. The associations between maltreatment and CVD were assessed using Poisson regression with robust variance to estimate risk ratios, stratified by sex and adjusted for socioeconomic and demographic factors.ResultsAll types of maltreatment were associated with increased risk of CVD and IHD in both sexes. Additionally, in women all types of maltreatment were associated with higher risk of hypertensive disease, and all, except emotional neglect, were associated with cerebrovascular disease. In men, all but sexual abuse, were associated with higher risk of hypertensive disease, and all, except physical and sexual abuse, were associated with cerebrovascular disease. Associations were generally stronger in women, and individuals who were younger at baseline had stronger associations of childhood maltreatment with any CVD and IHD, but age differences were less evident when only early onset CVD was considered.ConclusionsChildhood maltreatment was consistently associated with CVD and stronger associations were generally observed in women and seemed to be stronger for early onset CVD.

show abstract

Section: Strengths and Limitationsmentioning

confidence: 99%

Sex differences in the association between childhood maltreatment and cardiovascular disease in the UK Biobank

Soares

Hammerton

Howe

et al. 2020

Heart

View full text Add to dashboard Cite

show abstract

“…It cannot be generalized to all surgical procedures, and the results may differ for stable surgical teams. However, analyses based on stabilized IPT weighting reduce selection bias and make causal effects more plausible 40,59 . Effects of unmeasured co-variables and of ongoing developments, such as refined surgical procedures, cannot be ruled out.…”

Section: Discussionmentioning

confidence: 99%

Effects of structured intraoperative briefings on patient outcomes: multicentre before-and-after study

Tschan

Keller

Semmer

et al. 2021

British Journal of Surgery

View full text Add to dashboard Cite

Background Operations require collaboration between surgeons, anaesthetia professionals, and nurses. The aim of this study was to determine whether intraoperative briefings influence patient outcomes. Methods In a before-and-after controlled trial (9 months baseline; 9 months intervention), intraoperative briefings were introduced in four general surgery centres between 2015 and 2018. During the operation, the responsible surgeon (most senior surgeon present) briefed the surgical team using the StOP? protocol about: progress of the operation (Status), next steps (Objectives), possible problems (Problems), and encouraged asking questions (?). Differences between baseline and intervention were analysed regarding surgical-site infections (primary outcome), mortality, unplanned reoperations, and duration of hospital stay (secondary outcomes), using inverse probability of treatment (IPT) weighting based on propensity scores. Results In total, 8256 patients underwent surgery in the study. Endpoint data were available for 7745 patients (93.8 per cent). IPT-weighted and adjusted intention-to-treat analyses showed no differences in surgical-site infections between baseline and intervention (9.8 versus 9.6 per cent respectively; adjusted difference (AD) –0.15 (95 per cent c.i. −1.45 to 1.14) per cent; odds ratio (OR) 0.92, 95 per cent c.i. 0.83 to 1.15; P = 0.797), but there were reductions in mortality (1.6 versus 1.1 per cent; AD –0.54 (−1.04 to −0.03) per cent; OR 0.60, 0.39 to 0.92; P = 0.018), unplanned reoperations (6.4 versus 4.8 per cent; AD –1.66 (−2.69 to −0.62) per cent; OR 0.72, 0.59 to 0.89; P = 0.002), and fewer prolonged hospital stays (21.6 versus 19.8 per cent; AD –1.82 (−3.48 to −0.15) per cent; OR 0.87, 0.77 to 0.98; P = 0.024). Conclusion Short intraoperative briefings improve patient outcomes and should be performed routinely.

show abstract

“…Differences between survivor responders and non‐responders were evaluated using chi‐square tests or Fisher's exact tests as appropriate (Tables S1 and S2). To account for potential non‐response bias, demographic and treatment characteristics that differed between responders and non‐responders (Table S3) were incorporated into a propensity score through a logit model to determine the probability of participation 29 . The estimated propensity score was then used as a covariate in all regressions models.…”

Section: Methodsmentioning

confidence: 99%

Identifying clusters of health risk behaviors and their predictors in adult survivors of childhood cancer: A report from the French Childhood Cancer Survivor Study

et al. 2020

View full text Add to dashboard Cite

Objective: Health risk behaviors (HRB) of childhood cancer survivors (CCS) are generally studied separately, despite the evidence suggesting that HRB are not independent. To our knowledge, few studies have examined HRB profiles in the former pediatric cancer patients. In this study, we identified HRB profiles and examined predictors engaging in unhealthy behaviors in CCS. Methods: We used data from a French cohort of CCS that includes five-year survivors diagnosed between 1945 and 2000 and treated before reaching age 18, in five centers in France. A total of 2961 adult CCS answered a self-reported questionnaire pertaining to HRB. Latent class analysis was used to identify HRB profiles combining physical activity, smoking, cannabis use, and alcohol drinking. Multinomial logistic analyses examined predictors for engaging in unhealthy behaviors.

show abstract

Using Propensity Score Weighting to Reduce Selection Bias in Large-Scale Data Sets

Cited by 26 publications

References 33 publications

Sex differences in the association between childhood maltreatment and cardiovascular disease in the UK Biobank

Sex differences in the association between childhood maltreatment and cardiovascular disease in the UK Biobank

Effects of structured intraoperative briefings on patient outcomes: multicentre before-and-after study

Identifying clusters of health risk behaviors and their predictors in adult survivors of childhood cancer: A report from the French Childhood Cancer Survivor Study

Contact Info

Product

Resources

About