Introduction

Online data collection has become the standard for public health research due to its efficiency and increased participant access; however, studies have shown data quality suffers when using these methods. The purpose of this study was to determine differences in health outcome measures for good versus poor quality data after implementing rigorous data quality screening methods.

Methods

There were 2,315 responses collected from online survey platforms. Each record was categorized as good or poor quality based on 36 quality control measures. Demographics, outcomes, and psychometric properties were assessed. The outcome instrument, Foot and Ankle Activity Level Scale(FAALS), is intended for public health practitioners to assess activity level.

Results

For good-quality data (n=947), mean FAALS score=72.5 (SD=23.6), mean age=47.8 years (SD=17.3), 53.6% females/46.4% males, 10.15-minute completion, person reliability=0.90, item reliability=1.00. For poor-quality data (n=1368), mean FAALS score=55.5(SD=23.7), mean age=40.4 years (SD=15.0), 52.0% females/48.0% males, 8.84-minute completion, person reliability=0.92. There was a significant difference in mean FAALS scores for good vs poor quality data, t(2313)=-17.0, p<.001. Combined-quality data (n=2315), mean FAALS score=62.4 (SD=25.1), mean age=43.4 years (SD=16.4), 52.7% females/47.3% males, 9.48-minute completion, person reliability=0.91.

Conclusions

This study demonstrates an extremely large 17% difference in health outcomes for good vs poor quality data, while psychometrics were similar, emphasizing the need to integrate quality control measures into the online data collection process. Demographics also varied, presenting issues with representativeness and generalizability. By integrating quality control measures, public health practitioners can obtain more accurate results to inform policy and public health protocols.

Introduction. With the advent of big data, a plethora of analytical methods are required to address diverse issues. This study endeavors to examine the optimal analyses for leveraging big data in detecting factors associated with the care of acute ischemic stroke (AIS).

Method. We conducted a retrospective study involving consecutive patients admitted for AIS between 2014 and 2023 within a healthcare system comprising 14 hospitals. Demographic, socioeconomic, and clinical data were extracted from electronic health records. Dependent variables encompassed imaging utilization, treatment modalities, and discharge disposition. Our initial analysis employed multivariable logistic regressions, Lasso regressions, and the Random Forest method. Subsequently, we utilized GEE models to address the hierarchical data structure, conducted factor analysis for index creation, and employed cluster analysis to tackle heterogeneity issues.

Results. Out of the 1,981,132 admissions, 29,662 were for acute ischemic stroke (AIS). Among the AIS episodes, 32.4% occurred in individuals aged 80+, 50.3% female, and 20.6% Black. Additionally, 41.8% had diabetes, and 81.3% hypertension. Regarding imaging, 27.1% had CTA and 21.6% MRA. Among these cases, 36.6% had a LOS of 8+ days, and 16.1% resulted in death. Initial logistic regression analyses revealed disparities in various aspects. For instance, Black showed underuse of CTA (OR= 0.873, CI= [0.833,0.916]) and were associated with longer LOS (8+) (OR= 1.198, CI= [1.142,1.257]). However, further analyses employing different methods would obtain more accurate results.

Conclusion. It is imperative to recognize the importance of employing diverse analytical methods to effectively address the multitude of data situations encountered in research.

Background: The downstream-focused, specialty-centric US healthcare system and fragmented social services contribute to disparities and care navigation challenges for uninsured and underinsured populations, often high emergency department (ED) utilizers. NeighborhoodHELP, a household-centered approach by the Herbert Wertheim College of Medicine at Florida International University, provides comprehensive primary, preventive, behavioral health, and social services to underserved households through interprofessional student-resident-faculty-outreach teams, addressing social determinants of health (SDOH) and alleviating strain on EDs.

Methods: This study assessed NeighborhoodHELP's impact on avoidable ED utilization among uninsured frequent ED users referred to NeighborhoodHELP by Baptist Health South Florida (BHSF). Data about ED utilization was combined with NeighborhoodHELP program data. The New York University ED Algorithm classified visits as avoidable, unavoidable, or other using data from NeighborhoodHELP and BHSF. Participants with ≥1 avoidable ED visit within one year pre-enrollment and sufficient pre-post data were included. Percent change in ED reduction pre-post enrollment was examined. Poisson regression will assess factors related to ED utilization.

Results: Among 139 eligible participants (63% women, 48% Hispanic White, 31% non-Hispanic/Latino Black), avoidable ED visits decreased by 49.8% within one year post-enrollment, with further reductions of 64.6% (1-2 years) and 68.6% (2-3 years), highlighting NeighborhoodHELP's sustained impact on reducing avoidable ED visits over multiple years.

Conclusion: The significant, sustained decrease in avoidable ED visits for NeighborhoodHELP households emphasizes the potential of integrated community-based interventions, like those combining telehealth, mobile health, household visits, and community engagement, in reducing avoidable ED utilization, enhancing accessibility for underserved populations, and improving overall community health.