Improving Causal Inference: Recommendations for Covariate Selection and Balance in Propensity Score Methods

Kainz, Kirsten; Greifer, Noah; Givens, Ashley; Swietek, Karen; Lombardi, Brianna M.; Zietz, Susannah; Kohn, Jamie L.

doi:10.1086/691464

Cited by 41 publications

(51 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…An advantage of the PS analysis is that it allows the balance of covariates to be reviewed (e.g. using standardised differences) [ 35 , 36 ]. Achieving covariate balance strengthens the internal validity of our study and gives greater confidence that any effect observed may be due to the causal relationship between PCV13 and protection from hypoxic pneumonia [35] .…”

Section: Discussionmentioning

confidence: 99%

The effectiveness of the 13-valent pneumococcal conjugate vaccine against hypoxic pneumonia in children in Lao People's Democratic Republic: An observational hospital-based test-negative study

Weaver

Nguyen

Chan

et al. 2020

The Lancet Regional Health - Western Pacific

View full text Add to dashboard Cite

Background: Pneumococcal pneumonia is a leading cause of childhood mortality. Pneumococcal conjugate vaccines (PCVs) have been shown to reduce hypoxic pneumonia in children. However, there are no studies from Asia examining the effectiveness of PCVs on hypoxic pneumonia. We describe a novel approach to determine the effectiveness of the 13-valent PCV (PCV13) against hypoxia in children admitted with pneumonia in the Lao People's Democratic Republic. Methods: A prospective hospital-based, test-negative observational study of children aged up to 59 months admitted with pneumonia to a single tertiary hospital in Vientiane was undertaken over 54 months. Pneumonia was defined using the 2013 WHO definition. Hypoxia was defined as oxygen saturation < 90% in room air or requiring oxygen supplementation during hospitalisation. Test-negative cases and controls were children with hypoxic and non-hypoxic pneumonia, respectively. PCV13 status was determined by written record. Vaccine effectiveness was calculated using logistic regression. Propensity score and multiple imputation analyses were used to handle confounding and missing data. Findings: There were 826 children admitted with pneumonia, 285 had hypoxic pneumonia and 377 were PCV13-vaccinated. The unadjusted, propensity-score adjusted and multiple-imputation adjusted estimates of vaccine effectiveness against hypoxic pneumonia were 23% (95% confidence interval:-9, 46%; p = 0 • 14); 37% (6, 57%; p = 0 • 02) and 35% (7, 55%; p = 0 • 02) respectively. Interpretation: PCV13 is effective against hypoxic pneumonia in Asia, and should be prioritised for inclusion in national immunisation programs. This single hospital-based, test-negative approach can be used to assess vaccine effectiveness in other similar settings. Funding: Funded by the Bill & Melinda Gates Foundation.

show abstract

Section: Discussionmentioning

confidence: 99%

The effectiveness of the 13-valent pneumococcal conjugate vaccine against hypoxic pneumonia in children in Lao People's Democratic Republic: An observational hospital-based test-negative study

Weaver

Nguyen

Chan

et al. 2020

The Lancet Regional Health - Western Pacific

View full text Add to dashboard Cite

show abstract

“…In acute stroke, it is challenging to evaluate the effects of care on delirium independent of stroke severity because the need for care depends on it. Analysis using PS enables causal inference in observational studies by creating quasi-experimental comparability [24,27,29,32]. The results of this study thus provide evidence that some kinds of care are responsible for the development of delirium in patients with acute stroke and that reducing the amount of such care can prevent delirium.…”

Section: Discussionmentioning

confidence: 83%

“…In PS analysis, valid inference relies on a set of vital assumptions, no unmeasured confounding, and positivity (true PS should not be 0 or 1) [24,27,32]. We believe unmeasured confounding was minimal in this study because we prospectively measured a comprehensive set of known as well as potential predictors of delirium.…”

Section: Discussionmentioning

confidence: 99%

See 1 more Smart Citation

Effects of uncomfortable care and histamine H2-antagonists on delirium in acute stroke: A propensity score analysis

Nakamizo

Kanda²,

Kudo

et al. 2021

Journal of the Neurological Sciences

View full text Add to dashboard Cite

Background and purpose: Uncomfortable care and histamine H2 antagonist (H2A) are implicated in precipitating delirium. In acute stroke, however, the need for them depends on stroke severity, an established risk factor for delirium. So, it is unclear whether care or H2A itself is responsible for delirium. We aimed to evaluate their causal effects on delirium in acute stroke patients. Methods: This is a prospective cohort study on acute stroke patients admitted to a stroke care unit. Patients without stupor, coma, sedation, or delirium upon admission were enrolled. The treatment was H2A and five care modalities given during the first 24 h: restraint use, prohibited self-transfer, no oral feeding, indwelling catheters, and frequent nighttime care. The outcome was delirium within 5 days defined as Intensive Care Delirium Screening Checklist ≥4 points. We estimated the relative risk (RR) for delirium with regression models weighted by overlap weights using propensity scores estimated through logistic models incorporating known and potential confounders, including stroke severity. Results: Of the 387 participants, 188 were given at least one care modality and 130 were given H2A. A total of 42 developed delirium. Delirium was significantly associated with prohibited self-transfer (RR 1.7, 95% CI 1.0-3.0), frequent nighttime care (RR 2.1, 95% CI 1.2-3.7), and multiple care modalities (RR 2.4, 95% CI 1.3-4.4), while other care modalities and H2A were not. Conclusions: This study showed possible causal effects of uncomfortable care on delirium and suggests that minimizing it could prevent delirium in acute stroke.

show abstract

“…It is an effect size where the difference in covariate means between two groups is divided by the spread in means (Kainz et al 2017;Stuart 2010). The literature often refers to a threshold value of 0.1 (e.g., Haukoos and Lewis 2015;Kainz et al 2017). Standardised biases below this threshold indicate negligible imbalance whereas values exceeding this limit would point to considerable pre-event differences between both groups.…”

Section: Analytical Strategymentioning

confidence: 99%

Well-Being Before and After Pregnancy Termination: The Consequences of Abortion and Miscarriage on Satisfaction With Various Domains of Life

Huß

2021

J Happiness Stud

View full text Add to dashboard Cite

The consequences of pregnancy outcomes other than live birth on subjective well-being have rarely been analysed in research to date. This study examines pre-event determinants as well as the temporary and long-term effects of induced abortion and miscarriage (spontaneous abortion) on satisfaction with various domains of life. The data were derived from the German Panel Analysis of Intimate Relationships and Family Dynamics (pairfam). The longitudinal sample consists of 5331 women of reproductive age, of whom 214 women had an induced abortion, 331 women had a miscarriage, and 1156 women had a live birth during the observation period. First, pre-event measures of women who had an induced abortion and women who had a miscarriage were compared with the pre-event measures of those women who gave birth. Second, fixed effects models were used to examine whether overall or domain-specific life satisfaction changed following a pregnancy termination. The results show that pregnancies resulting in abortion or miscarriage were less frequently preceded by pregnancy intentions compared to those resulting in live birth, and that induced abortion—but not miscarriage—was furthermore accompanied by lower pre-event satisfaction than live birth. Following both miscarriage and induced abortion, women experienced temporary declines in overall life satisfaction and showed persistently lower satisfaction in several domains of life. With regard to induced abortion, pre-event measures were a better predictor of overall well-being than the consequences of the event itself. Low life satisfaction might therefore be a risk factor for having an abortion rather than a result.

show abstract

Improving Causal Inference: Recommendations for Covariate Selection and Balance in Propensity Score Methods

Cited by 41 publications

References 41 publications

The effectiveness of the 13-valent pneumococcal conjugate vaccine against hypoxic pneumonia in children in Lao People's Democratic Republic: An observational hospital-based test-negative study

The effectiveness of the 13-valent pneumococcal conjugate vaccine against hypoxic pneumonia in children in Lao People's Democratic Republic: An observational hospital-based test-negative study

Effects of uncomfortable care and histamine H2-antagonists on delirium in acute stroke: A propensity score analysis

Well-Being Before and After Pregnancy Termination: The Consequences of Abortion and Miscarriage on Satisfaction With Various Domains of Life

Contact Info

Product

Resources

About