Misleading Epidemiological and Statistical Evidence in the Presence of Simpson's Paradox: An Illustrative Study Using Simulated Scenarios of Observational Study Designs

A scoping review shows that no single existing risk of bias assessment tool considers all sources of bias for cross-sectional studies

OBJECTIVES

Different tools to assess the potential risk of bias (RoB) for cross-sectional studies have been developed, but it is unclear whether all pertinent bias concepts are addressed. We aimed to identify RoB concepts applicable to cross-sectional research validity and to explore coverage for each in existing appraisal tools.

STUDY DESIGN AND SETTING

This scoping review followed the Joanna Briggs Institute methodology. We included records of any study design describing or reporting methods, concepts or tools used to consider RoB in health research reported to be descriptive/prevalence survey or analytic/association (cross-sectional) study designs. Synthesis included quantitative and qualitative analysis.

RESULTS

Of the 4556 records screened, 90 were selected for inclusion; 67 (74%) described the development of, or validation process for, appraisal tools, 15 (17%) described methodological content or theory relevant to RoB for cross-sectional studies and 8 (9%) records of methodological systematic reviews. Review of methodological reports identified important RoB concepts for both descriptive/prevalence and analytic/association studies. Tools identified (n = 64 unique tools) were either intended to appraise quality or assess RoB in multiple study designs including cross-sectional studies (n = 21; 33%) or cross-sectional designs alone (n = 43; 67%). Several existing tools were modified (n = 17; 27%) for application to cross-sectional studies. The RoB items most frequently addressed in the RoB tools were validity and reliability of the exposure (53%) or outcome (65%) measurement and representativeness of the study population (59%). Most tools did not consider nonresponse or missingness appropriately or at all.

CONCLUSION

Assessing cross-sectional studies involve unique RoB considerations. We identified RoB tools designed for broad applicability across various study designs as well as those specifically tailored for cross-sectional studies. However, none of the identified tools comprehensively address all potential biases pertinent to cross-sectional studies. Our findings indicate a need for continued improvement of RoB tools and suggest that the development of context-specific or more precise tools for this study design may be necessary.

Outcomes after early treatment with hydroxychloroquine and azithromycin: An analysis of a database of 30,423 COVID-19 patients

Background

Many studies have evaluated the use of hydroxychloroquine in COVID-19. Most retrospective observational studies demonstrate a benefit of using HCQ on mortality, but not most randomized clinical trials.

Methods

We analyzed raw data collected from a cohort of 30,423 patients with COVID-19 cared for at IHU Méditerranée Infection in Marseille France and extracted from the DRYAD open data platform. We performed univariate and multivariable logistic regressions with all-cause mortality within six weeks. Multivariable logistic regressions were adjusted for sex, age group (<50, 50-69, 70-89 and ​> ​89 years), periods (or variants), and type of patient management.

Results

Among 30,202 patients for whom information on treatment was available, 191/23,172 (0.82%) patients treated with HCQ-AZ died, compared to 344/7030 (4.89%) who did not receive treatment with HCQ-AZ. HCQ-AZ therapy was associated with a lower mortality than treatment without HCQ-AZ (odds ratio (OR) 0.16; 95% confidence interval (CI), 0.14-0.19). After adjustment for sex, age, period, and patient management, HCQ-AZ was associated with a significantly lower mortality rate (adjusted OR (aOR) 0.55, 95% CI 0.45-0.68). On a subsample of 21,664 patients with available variant information, results remained robust after adjustment on sex, age, patient management and variant (aOR 0.55; 95% CI 0.44-0.69). On a subsample of 16,063 patients, HCQ-AZ was still associated with a significantly lower mortality rate (aOR 0.47, 95%CI 0.29-0.75) after adjustment for sex, age, period, patient management, vaccination status and comorbidities.

Conclusion

Analysis of this large online database showed that HCQ-AZ was consistently associated with the lowest mortality.

Age and CD4+ T cell counts are inversely associated with HIV drug resistance mutations in treatment naive female sex workers

The increasing prevalence of human immunodeficiency virus (HIV) drug resistance mutations (HIVDRM) in untreated seropositive persons has consequences for future treatment options. This is extremely important in key populations such as female sex workers (FSWs), where the prevalence of pretreatment drug resistance (PDR) and associated risk factors are unknown. In this study, we analyzed PDR and associated risk factors in recently diagnosed and treatment-naive FSWs in Nairobi, Kenya. In this cross-sectional study, we used 64 HIV-seropositive plasma samples collected from FSWs between November 2020 and April 2021. To identify HIVDRM, the pol gene was amplified and genotyped using sanger sequencing. The effects of age, tropism, CD4+ T cell count, subtype, and location on HIVDRM counts were examined using Poisson regression. Overall, the prevalence of PDR was 35.9% (95% CI: 24.3-48.9), which was strongly influenced by K103N and M184V mutations, which confer resistance to non-nucleoside reverse transcriptase inhibitors (NNRTIs) and nucleoside reverse transcriptase inhibitors (NRTI), respectively. Subtype A1 was predominant followed by subtype D with a notable increase in inter-subtype recombinants. We found statistically significant evidence that age was inversely related to HIVDRM. A FSW who is 1 year older had 12% less HIVDRM (incidence rate ratios [IRR]: 0.88; 95% CI: 0.82-0.95; P < .001), after adjusting for CD4+ T cell count, subtype, location, and tropism. Similarly, an increase in CD4+ T cell count by 1 unit, was associated with 0.4% fewer HIVDRM (IRR: 0.996; 95% CI: 0.994-0.998; P = .001), while controlling for the other variables. HIV-1 tropism was not associated with HIVDRM counts. In conclusion, our findings show a high prevalence of NNRTIs. Lower CD4+ T cell counts and younger age were significant risk factors that influenced HIVDRM loads. This finding underscores the relevance of targeted interventions and the importance of continuing to focus on FSWs as a way of addressing the HIV epidemic.

Net charge and position 22 of the V3 loop are associated with HIV-1 tropism in recently infected female sex workers in Nairobi, Kenya

Human immunodeficiency virus (HIV) infection affects around 37 million people worldwide, and in Kenya, key populations especially female sex workers (FSW), are thought to play a substantial role in the wider, mostly heterosexual HIV-1 transmission structure. Notably, HIV tropism has been found to correlate with HIV-1 transmission and disease progression in HIV-infected patients. In this study, recently infected FSWs from Nairobi, Kenya, were assessed for HIV tropism and the factors related to it. We used a cross-sectional study design to analyze 76 HIV-1 positive plasma samples obtained from FSWs enrolled in sex worker outreach program clinics in Nairobi between November 2020 and April 2021. The effects of clinical, demographic, and viral genetic characteristics were determined using multivariable logistic regression. HIV-1 subtype A1 accounted for 89.5% of all cases, with a prevalence of CXCR4-tropic viruses of 26.3%. WebPSSMR5X4 and Geno2Pheno [G2P:10-15% false positive rate] showed high concordance of 88%. Subjects infected with CXCR4-tropic viruses had statistically significant lower baseline CD4+T-cell counts than those infected with CCR5-tropic viruses (P = .044). Using multivariable logistic regression and adjusting for potential confounders, we found that net charge, the amino acid at position 22 of the V3 loop, and the geographic location of the subject were associated with tropism. A unit increase in V3 loop's net-charge increased the odds of a virus being CXCR4-tropic by 2.4 times (OR = 2.40, 95%CI = 1.35-5.00, P = .007). Second, amino acid threonine at position 22 of V3 loop increased the odds of a strain being X4 by 55.7 times compared to the alanine which occurred in CCR5-tropic strains (OR = 55.7, 95%CI = 4.04-84.1, P < .003). The Kawangware sex worker outreach program clinic was associated with CXCR4-tropic strains (P = .034), but there was there was no evidence of a distinct CXCR4-tropic transmission cluster. In conclusion, this study revealed a high concordance of WebPSSMR5X4 and Geno2Pheno in predicting HIV tropism. The most striking finding was that amino acid position 22 of the V3 loop is linked to tropism in HIV-1 subtype A1. Additional studies with a large dataset are warranted to confirm our findings.

Spatial statistical machine learning models to assess the relationship between development vulnerabilities and educational factors in children in Queensland, Australia

BACKGROUND

The health and development of children during their first year of full time school is known to impact their social, emotional, and academic capabilities throughout and beyond early education. Physical health, motor development, social and emotional well-being, learning styles, language and communication, cognitive skills, and general knowledge are all considered to be important aspects of a child's health and development. It is important for many organisations and governmental agencies to continually improve their understanding of the factors which determine or influence development vulnerabilities among children. This article studies the relationships between development vulnerabilities and educational factors among children in Queensland, Australia.

METHODS

Spatial statistical machine learning models are reviewed and compared in the context of a study of geographic variation in the association between development vulnerabilities and attendance at preschool among children in Queensland, Australia. A new spatial random forest (SRF) model is suggested that can explain more of the spatial variation in data than other approaches.

RESULTS

In the case study, spatial models were shown to provide a better fit compared to models that ignored the spatial variation in the data. The SRF model was shown to be the only model which can explain all of the spatial variation in each of the development vulnerabilities considered in the case study. The spatial analysis revealed that the attendance at preschool factor has a strong influence on the physical health domain vulnerability and emotional maturity vulnerability among children in their first year of school.

CONCLUSION

This study confirmed that it is important to take into account the spatial nature of data when fitting statistical machine learning models. A new spatial random forest model was introduced and was shown to explain more of the spatial variation and provide a better model fit in the case study of development vulnerabilities among children in Queensland. At small-area population level, increased attendance at preschool was strongly associated with reduced physical and emotional development vulnerabilities among children in their first year of school.

The quantification of Simpson's paradox and other contributions to contingency table theory

The analysis of contingency tables is a powerful statistical tool used in experiments with categorical variables. This study improves parts of the theory underlying the use of contingency tables. Specifically, the linkage disequilibrium parameter as a measure of two-way interactions applied to three-way tables makes it possible to quantify Simpson’s paradox by a simple formula. With tests on three-way interactions, there is only one that determines whether the partial interactions of all variables agree or whether there is at least one variable whose partial interactions disagree. To date, there has been no test available that determines whether the partial interactions of a certain variable agree or disagree, and the presented work closes this gap. This work reveals the relation of the multiplicative and the additive measure of a three-way interaction. Another contribution addresses the question of which cells in a contingency table are fixed when the first- and second-order marginal totals are given. The proposed procedure not only detects fixed zero counts but also fixed positive counts. This impacts the determination of the degrees of freedom. Furthermore, limitations of methods that simulate contingency tables with given pairwise associations are addressed.

The UK foot and ankle COVID-19 national (FAlCoN) audit - Regional variations in COVID-19 infection and national foot and ankle surgical activity

AIMS

This paper details the impact of COVID-19 on foot and ankle activity in the UK. It describes regional variations and COVID-19 infection rate in patients undergoing foot and ankle surgery before, during and after the first national lock-down.

PATIENTS & METHODS

This was a multicentre, retrospective, UK-based, national audit on foot and ankle patients who underwent surgery between 13th January and 31st July 2020. Data was examined pre- UK national lockdown, during lockdown (23rd March to 11th May 2020) and post-lockdown. All adult patients undergoing foot and ankle surgery in an operating theatre during the study period included from 43 participating centres in England, Scotland, Wales and Northern Ireland. Regional, demographic and COVID-19 related data were captured.

RESULTS

6644 patients were included. In total 0.53% of operated patients contracted COVID-19 (n = 35). The rate of COVID-19 infection was highest during lockdown (2.11%, n = 16) and lowest after lockdown (0.16%, n = 3). Overall mean activity during lockdown was 24.44% of pre-lockdown activity with decreases in trauma, diabetic and elective foot and ankle surgery; the change in elective surgery was most marked with only 1.73% activity during lock down and 10.72% activity post lockdown as compared to pre-lockdown. There was marked regional variation in numbers of cases performed, but the proportion of decrease in cases during and after lockdown was comparable between all regions. There was also a significant difference between rates of COVID-19 and timing of peak, cumulative COVID-19 infections between regions with the highest rate noted in South East England (3.21%). The overall national peak infection rate was 1.37%, occurring during the final week of lockdown. General anaesthetic remained the most common method of anaesthesia for foot and ankle surgery, although a significant increase in regional anaesthesia was witnessed in the lock-down and post-lockdown periods.

CONCLUSIONS

National surgical activity reduced significantly for all cases across the country during lockdown with only a slow subsequent increase in elective activity. The COVID-19 infection rate and peaks differed significantly across the country.

A cohort study of hepatectomy-related complications and prediction model for postoperative liver failure after major liver resection in 1,441 patients without obstructive jaundice

Background

This cohort study, based on a large sample of extensive hepatectomy cases, aimed to analyze the distribution of hepatectomy-related complications and to develop a predictive model of posthepatectomy liver failure (PHLF).

Methods

Data of patients who underwent hepatectomy of ≥3 liver segments at the Eastern Hepatobiliary Surgery Hospital from 2000 to 2016 were collected and analyzed. Information on hepatectomy-related complications was collected and risk factors were analyzed. A total of 1,441 eligible patients were randomly assigned at 3:1 ratio into the derivation (n=1,080) and validation (n=361) cohorts. The multivariable logistic regression model was used to establish the prediction model of PHLF in the derivation cohort.

Results

The incidence rates of PHLF, ascites, bile leakage, intra-abdominal bleeding, and abscesses were 58.22%, 10.76%, 11.17%, 9.71%, and 4.16%, respectively. The 90-day perioperative mortality rate was 1.32%. Multivariate analyses found that age, gender, platelet, creatinine, gamma-glutamyltransferase, thrombin time, fibrinogen, hepatitis B e (HBe) antigen positive, and number of resected liver segments were independent prognostic factors of PHLF in the derivation cohort and included in the nomogram. The prediction model demonstrated good discrimination [area under the curve =0.726, 95% confidence interval (CI), 0.696–0.760, P<0.0001] and calibration.

Conclusions

Our study showed a high perioperative safety and a low risk of serious complications in patients who underwent major liver resection (MLR) at a large hepatobiliary surgery center. Routine preoperative clinical information can be used to develop a postoperative liver failure risk prediction model for rational planning of surgery.