A tutorial comparing different covariate balancing methods with an application evaluating the causal effects of substance use treatment programs for adolescents

Markoulidakis, Andreas; Taiyari, Khadijeh; Holmans, Peter; Pallmann, Philip; Busse, Monica; Godley, Mark D.; Griffin, Beth Ann

doi:10.1007/s10742-022-00280-0

Cited by 19 publications

(13 citation statements)

References 68 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Therefore, the threshold for imbalance ought to become less stringent as sample size falls. On the second, we agree that a threshold is arbitrary at first, but the observational research field including most of these authors have come to relative agreement on a 0.1 constant threshold [3][4][5][6][7][8][9][10][11][12][13][14][15] despite the fact that no threshold can guarantee immunity from important bias. The appropriateness of a threshold is decided slowly over time as real study results are compared to baseline knowledge and validated in later randomized experiments.…”

Section: Discussionmentioning

confidence: 57%

Assessing Covariate Balance with Small Sample Sizes

Hripcsak,

Zhang,

et al. 2024

Preprint

View full text Add to dashboard Cite

Propensity score adjustment addresses confounding by balancing covariates in subject treatment groups through matching, stratification, inverse probability weighting, etc. Diagnostics ensure that the adjustment has been effective. A common technique is to check whether the standardized mean difference for each relevant covariate is less than a threshold like 0.1. For small sample sizes, the probability of falsely rejecting the validity of a study because of chance imbalance when no underlying balance exists approaches 1. We propose an alternative diagnostic that checks whether the standardized mean difference statistically significantly exceeds the threshold. Through simulation and real-world data, we find that this diagnostic achieves a better trade-off of type 1 error rate and power than standard nominal threshold tests and not testing for sample sizes from 250 to 4000 and for 20 to 100,000 covariates. In network studies, meta-analysis of effect estimates must be accompanied by meta-analysis of the diagnostics or else systematic confounding may overwhelm the estimated effect. Our procedure for statistically testing balance at both the database level and the meta-analysis level achieves the best balance of type-1 error rate and power. Our procedure supports the review of large numbers of covariates, enabling more rigorous diagnostics.

show abstract

Section: Discussionmentioning

confidence: 57%

Assessing Covariate Balance with Small Sample Sizes

Hripcsak,

Zhang,

et al. 2024

Preprint

View full text Add to dashboard Cite

show abstract

“…We addressed these measured and unmeasured confounding variables using do‐ calculus logic (Pearl, 1995, 2009; Shrier & Platt, 2008; Suttorp et al., 2015) (see Supporting Information for details). EBAL weights were assigned to the pixels to ensure that pixels of the four different vegetation classes were comparable with respect to the four topographical covariates and distance to fire centroid (Greifer, 2019; Markoulidakis et al., 2022; Matschinger et al., 2020). Using the weighted data, we built a logistic regression model that predicted fire susceptibility from vegetation type, topographical variables and distance from burn area centroid.…”

Section: Methodsmentioning

confidence: 99%

“…Additionally, new statistical tools have also made it easier to handle data extracted from burn area and vegetation time series. There are now standardised workflows to handle covariate imbalance, such as the tendency for forests to be found in wetter valleys, by matching or reweighting (Cannas & Arpino, 2019; Markoulidakis et al., 2022). Survival analysis, which is used to model fire‐vegetation feedbacks (Reed et al., 1998; Tepley et al., 2018) and vegetation succession (Longpre & Morris, 2012), has also developed rapidly.…”

Section: Introductionmentioning

confidence: 99%

Fire traps in the wet subtropics: New perspectives from Hong Kong

Chan,

Coomes

2024

Journal of Applied Ecology

View full text Add to dashboard Cite

Fires undermine efforts to restore degraded forests in the wet tropics and subtropics. Grasslands and shrublands established after fires are more fire susceptible than forests and tend to be set alight more often, creating a positive feedback loop that curbs succession (i.e. create a fire trap). Understanding the factors that underpin the strength of these fire traps could transform restoration programmes by identifying the steps needed to escape them. Fire traps are notoriously challenging to quantify because multiple factors influence fire occurrence and vegetation recovery. Here we used multi‐decadal satellite imagery from Landsat to create a 34‐year time series of burn areas and vegetation dynamics in wet subtropical Hong Kong. These dynamic maps were then used to characterise (1) the influence of successional stage on fire occurrence, having accounted for topographical and ignition source imbalances using neighbourhood analyses and entropy balancing (EBAL) weights, and (2) recovery time to the next successional stage by survival analysis with EBAL weights. Our analyses revealed that fire regimes in the wet subtropics are defined by strong fire‐vegetation feedbacks. Grasslands and shrublands were 20 and 9 times more susceptible to fires than forests in similar topographic positions. Human activities compounded these differences by disproportionally introducing more ignition sources to grasslands (2.3 times) and shrublands (2.0 times) than to forests. Burnt shrublands recovered to forests faster (19 years) than grasslands (40 years). Proximity to forest patches had strong positive effects on recovery rates, highlighting the importance of seed sources. Post‐fire recovery was faster on wetter northwest‐facing sites and valleys. Overall, topography strongly influenced recovery processes but hardly affected fire occurrence. Synthesis and applications. Our study provides the first quantification of fire trap processes in the wet subtropics, which provides new opportunities for evidence‐based fire suppression and post‐fire restoration. Our results suggest that (1) fire traps could be mitigated by fire suppression programmes as the traps are currently exacerbated by ignition source imbalance; (2) establishing green fire breaks represents an effective fire suppression measure in the wet subtropics; and (3) active restoration could target areas where models predict sluggish post‐fire natural regeneration.

show abstract

“…The SMD reports the magnitude of difference between groups, and is independent of the sample size. SMD>.1 (10%) was interpreted as a meaningful effect size 17,18 . We compared donor characteristics across AKI stages using Pearson's chi‐squared test for categorical variables and Cuzick's test based on ranks for continuous variables.…”

Section: Methodsmentioning

confidence: 99%

Kidney nonprocurement in deceased donors with acute kidney injury

Husain

King

et al. 2022

Clinical Transplantation

View full text Add to dashboard Cite

Background: Acute kidney injury (AKI) is common in deceased organ donors and is associated with high rates of kidney discard by transplant centers. High discard rates may consequently drive nonprocurement of these kidneys by organ procurement organizations. We aimed to study the relationship between donor AKI and kidney nonprocurement.Methods: Using U.S. registry data, we identified donors with at least one organ recovered from 2008 to 2018. We compared characteristics of donors with no kidneys procured across AKI stages, and used multivariable logistic regression to evaluate the relationship between AKI severity and kidney nonprocurement.Results: Overall 14 543 kidneys from 7620 donors were not procured, among which 93% were from donors with AKI. For 6945 donors with no kidneys procured but an extrarenal organ recovered, most had stage 3 (51%), followed by stage 1 (27%) and stage 2 AKI (15%). Nonprocured stage 3 donors were the youngest and had the lowest Kidney Donor Risk Index of all nonprocured donors. Adjusted odds of kidney nonprocurement were 1.14 (95%CI 1.02-1.27) for stage 1, 1.25 (95%CI 1.12-1.41) for stage 2, and 10.37 (95%CI 9.30-11.56) for stage 3 donors, compared to non-AKI donors.Among donors with minimum creatinine <1.5 mg/dl, stage 2 and 3 AKI were still associated with significantly higher odds of nonprocurement.Conclusions: AKI severity is a strong risk factor for kidney nonprocurement. Efforts to address the organ shortage should focus on encouraging procurement and utilization of kidneys from deceased donors with severe AKI, given the large and rising prevalence of donor AKI and excellent transplant outcomes with these kidneys.

show abstract

A tutorial comparing different covariate balancing methods with an application evaluating the causal effects of substance use treatment programs for adolescents

Cited by 19 publications

References 68 publications

Assessing Covariate Balance with Small Sample Sizes

Assessing Covariate Balance with Small Sample Sizes

Fire traps in the wet subtropics: New perspectives from Hong Kong

Kidney nonprocurement in deceased donors with acute kidney injury

Contact Info

Product

Resources

About