Ronnie Pingel

Effects of correlated covariates on the asymptotic efficiency of matching and inverse probability weighting estimators for causal inference

In observational studies, the overall aim when fitting a model for the propensity score is to reduce bias for an estimator of the causal effect. To make the assumption of an unconfounded treatment plausible researchers might include many, possibly correlated, covariates in the propensity score model. In this paper, we study how the asymptotic efficiency of matching and inverse probability weighting estimators for average causal effects change when the covariates are correlated. We investigate the case with multivariate normal covariates, a logistic model for the propensity score and linear models for the potential outcomes and show results under different model assumptions. We show that the correlation can both increase and decrease the large sample variances of the estimators, and that the correlation affects the asymptotic efficiency of the estimators differently, both with regard to direction and magnitude. Moreover, the strength of the confounding towards the outcome and the treatment plays an important role.

Pelvic pain after childbirth: a longitudinal population study.

Abstract In this longitudinal population study, the aims were to study associations of mode of delivery with new onset of pelvic pain and changes in pelvic pain scores up to 7 to 18 months after childbirth. We included 20,248 participants enrolled in the Norwegian Mother and Child Cohort Study (1999-2008) without preexisting pelvic pain in pregnancy. Data were obtained by 4 self-administered questionnaires and linked to the Medical Birth Registry of Norway. A total of 4.5% of the women reported new onset of pelvic pain 0 to 3 months postpartum. Compared to unassisted vaginal delivery, operative vaginal delivery was associated with increased odds of pelvic pain (adjusted odds ratio [OR]: 1.30; 95% confidence interval [CI]: 1.06-1.59). Planned and emergency cesarean deliveries were associated with reduced odds of pelvic pain (adjusted OR: 0.48; 95% CI: 0.31-0.74 and adjusted OR: 0.65; 95% CI: 0.49-0.87, respectively). Planned cesarean delivery, young maternal age, and low Symptom Checklist-8 scores were associated with low pelvic pain scores after childbirth. A history of pain was the only factor associated with increased pelvic pain scores over time (P = 0.047). We conclude that new onset of pelvic pain after childbirth was not commonly reported, particularly following cesarean delivery. Overall, pelvic pain scores were rather low at all time points and women with a history of pain reported increased pelvic pain scores over time. Hence, clinicians should follow up women with pelvic pain after a difficult childbirth experience, particularly if they have a history of pain.

Psychological mediators related to clinical outcome in cognitive behavioural therapy for coronary heart disease: A sub-analysis from the SUPRIM trial:

Background The Secondary Prevention in Uppsala Primary Healthcare Project (SUPRIM) was a randomized controlled trial of a group-based cognitive behavioural therapy stress management programme for patients with coronary heart disease. The project was successful in reducing the risk of fatal or non-fatal first recurrent cardiovascular events. The aim of this study was to analyse the effect of cognitive behavioural therapy on self-rated stress, somatic anxiety, vital exhaustion and depression and to study the associations of these factors with the reduction in cardiovascular events. Methods A total of 362 patients were randomly assigned to intervention or usual care groups. The psychological outcomes were assessed five times during 24 months and analysed using linear mixed models. The mediating roles of the outcomes were analysed using joint modelling of the longitudinal and time to event data. Results The intervention had a positive effect on somatic anxiety (p < 0.05), reflecting a beneficial development over time compared with the controls. Stress, vital exhaustion and depression did not differ between the groups over time. Mediator analysis suggested that somatic anxiety may have mediated the effect of treatment on cardiovascular events. Conclusions The intervention had a small positive effect on somatic anxiety, but did not affect stress, vital exhaustion or depression in patients with coronary heart disease. Somatic anxiety was associated with an increased risk of cardiovascular events and might act as a partial mediator in the treatment effect on cardiovascular events. However, the mechanisms between the intervention and the protective cardiovascular outcome remain to be identified.

Education, immigration and income as risk factors for hemoglobin A1c >70 mmol/mol when diagnosed with type 2 diabetes or latent autoimmune diabetes in adult: a population-based cohort study

Objectives The aim of this research is to study education, income and immigration as risk factors for high hemoglobin A1c (HbA1c >70 mmol/mol (8.6%)) when diagnosed with type 2 diabetes (T2D) or latent autoimmune diabetes in the adult (LADA). Research design and methods Patients were included from the All New Diabetics in Scania study (2008-2013). Level of education, disposable income and immigration year were retrieved from the longitudinal integrated database for labour market research (LISA) register compiled by Statistics Sweden. Logistic regression models were used to estimate ORs for HbA1c >70 mmol/mol (8.6%) at diagnosis. Results A total of 3794 patients with incident T2D (n=3 525) or LADA (n=269) were included. Patients with T2D with a low (≤9 years) or medium (10-12 years) levels of education were more likely to have high HbA1c at diagnosis compared with patients with T2D with a high (>12 years) level of education (OR 1.34, 95% CI 1.08 to1.66, OR 1.26, 95% CI 1.03 to 1.54). Low-income patients with T2D (<60% of median) were more likely to have high HbA1c at diagnosis compared with high-income patients withT2D (>150% of median) (OR 1.35, 95% CI 1.02 to 1.79). Conclusions Patients with lower levels of education or low income and are more likely to have HbA1c is >70 mmol/mol (8.6%) when diagnosed with T2D. An understanding of how socioeconomic position influences the clinical presentation at diagnosis may facilitate screening programs designed to target populations at risk for delayed diagnosis.

The interdependence of posttraumatic stress symptoms in parental dyads during and after their child's treatment for cancer

Abstract Background: Cancer in a child is highly distressing and some parents are at increased risk for developing posttraumatic stress symptoms (PTSS). However, the interdependence of PTSS in parental dyads has rarely been accounted for. The aim was to explore the dyadic relationship of PTSS in parents of children diagnosed with cancer. Material and methods: The sample includes 150 parents (75 dyads) of 75 children diagnosed with cancer in Sweden during 2002–2004, with follow-up until one year after end of treatment. Data on PTSS from six assessments were included. The first three assessments were carried out during treatment and the remaining after end of treatment. Actor-partner interdependence models were estimated using a structural equation modeling approach to explore the dyadic relationship of PTSS. Actor effects refer to intra-individual dependency over time, and partner effects refer to inter-individual dependency over time, i.e., how much an individual’s symptom levels are affected by their partner’s symptom levels at the previous assessment. Results: Results show both actor and partner effects during the child’s treatment. Only an actor effect remained following end of treatment where level of PTSS at one assessment was associated with the level of PTSS at the subsequent assessment. The association between mothers’ and fathers’ PTSS did not remain after end of treatment. Conclusions: Parents appear to react as an interdependent emotional system during the child’s treatment but this effect disappears after end of treatment. Results suggest psychological interventions for parents during the child’s cancer treatment should also be sensitive to and address the influence that distress in one partner may have on the other.

The effect of group-based cognitive behavioral therapy on inflammatory biomarkers in patients with coronary heart disease—results from the SUPRIM-trial

Abstract Background: The Secondary Prevention in Uppsala Primary Healthcare Project (SUPRIM) is a prospective randomized controlled trial of a group-based cognitive behavioral therapy (CBT) stress management program for coronary heart disease (CHD) patients. The intervention reduced the risk of fatal or non-fatal first recurrent cardiovascular (CV) events. The aim of the present study was to analyze if the positive effects of the CBT program on clinical outcomes could have been mediated by changes in biomarkers for inflammation. Methods: Altogether 362 patients with CHD were randomly assigned to intervention or usual care. The inflammatory biomarkers (VCAM-1, TNF-R1, TNF-R2, PTX3, and hs-CRP) were serially assessed at five time points every six months from study start until 24 months later, and analyzed with linear mixed models. Results: Baseline levels of the inflammatory markers were near normal, indicating a stable phase. The group-based CBT stress management program did not significantly affect the levels of inflammatory biomarkers in patients with CHD. Three out of five (VCAM-1, TNF-R2, and PTX3) inflammatory biomarkers showed a slight increase over time in both study groups, and all were positively associated with age. Conclusion: Group-based CBT stress management did not affect biomarkers for inflammation in patients with CHD. It is therefore unlikely that inflammatory processes including these biomarkers were mediating the effect the CBT program had on the reduction in CV events. The close to normal baseline levels of the biomarkers and the lack of elevated psychological distress symptoms indicate a possible floor effect which may have influenced the results.

The relationship between self-efficacy and transition to work or studies in young adults with disabilities:

Aim: To investigate perceived self-efficacy in unemployed young adults with disabilities, and the association between self-efficacy and transition to work or studies. Methods: This prospective cohort study collected data through self-report questionnaires and registry data from a vocational rehabilitation project with young adults, aged 19–29 years. The Swedish Social Insurance Agency, the Swedish Public Employment Service and the participating municipalities identified potential participants for the study. A total of 531 participants were included in the study, of which 249 (47%) were available for analysis. Multinomial logistic regression models were carried out to estimate the associations between self-efficacy, demographic (age, country of birth, education level), health and employment status. The latter was coded as: ‘no transition to work or studies’, ‘transition to studies’, and ‘transition to work’. Results: A higher level of self-efficacy was associated with increased odds for ‘transition to work’ (OR = 2.37, p < 0.05). This finding remained consistent when adjusting for possible confounders. The mean value of self-efficacy was low, and participants with lower self-efficacy reported worse self-rated health (p < 0.001) compared to participants with higher self-efficacy. Conclusions: The results from this study suggest that self-efficacy should be addressed in the vocational rehabilitation of young adults with disabilities in order to support their transition and integration into the labour market.

Risk selection in primary care – a cross-sectional fixed effect analysis of Swedish individual data

Objective To assess socioeconomic differences between patients registered with private and public primary healthcare centres. Design Population-based cross-sectional study controlling for municipality and household. Setting Swedish population-based socioeconomic data collected from Statistics Sweden linked with individual registration data from all 21 Swedish regions. Participants All individuals residing in Sweden on 31 December 2015 (n=9 851 017) were included in the study. Primary outcome measures Registration with private versus public primary healthcare centres. Results After controlling for municipality and household, individuals with higher socioeconomic status were more likely to be registered with a private primary healthcare provider. Individuals in the highest income quantile were 4.9 percentage points (13.7%) more likely to be registered with a private primary healthcare provider compared with individuals in the lowest income quantile. Individuals with 1–3 years of higher education were 4.7 percentage points more likely to be registered with a private primary healthcare provider compared with those with an incomplete primary education. Conclusions The results show that there are notable differences in registration patterns, indicating a skewed distribution of patients and health risks between private and public primary healthcare providers. This suggests that risk selection behaviour occurs in the reformed Swedish primary healthcare system, foremost through location patterns.

Correlation and efficiency of propensity score-based estimators for average causal effects

ABSTRACT Propensity score-based estimators are commonly used to estimate causal effects in evaluation research. To reduce bias in observational studies, researchers might be tempted to include many, perhaps correlated, covariates when estimating the propensity score model. Taking into account that the propensity score is estimated, this study investigates how the efficiency of matching, inverse probability weighting, and doubly robust estimators change under the case of correlated covariates. Propositions regarding the large sample variances under certain assumptions on the data-generating process are given. The propositions are supplemented by several numerical large sample and finite sample results from a wide range of models. The results show that the covariate correlations may increase or decrease the variances of the estimators. There are several factors that influence how correlation affects the variance of the estimators, including the choice of estimator, the strength of the confounding toward outcome and treatment, and whether a constant or non-constant causal effect is present.