Distributional and heterogeneous treatment effects of health care interventions for gestational diabetes using regression discontinuity in electronic health record data

Date of Approval: 
2020-12-09 00:00:00
Lay Summary: 
This study analyses the effects of health care interventions for maternal and child outcomes for women diagnosed with gestational diabetes. Gestational diabetes affects health outcomes of two individuals at the same time and its prevalence is increasing due to an increase in older and more overweight or obese women becoming pregnant. Therefore, studying effects of interventions is of high relevance for clinical care. While clinical trials have analysed outcomes around the time of birth, they often lack a long-term perspective and may thus not be able to fully assess adverse outcomes for the mother or child later in life such as the onset of type 2 diabetes. On the other hand, observational studies provide real-life insights and can be collected over a longer time period but face challenges in establishing causality. This study exploits random variation in treatment decisions based on major clinical guidelines in the UK in order to test the real-life effectiveness of gestational diabetes interventions. To this end, we apply a regression discontinuity design which compares the outcomes for individuals just above and below a threshold to diagnose gestational diabetes. The large number of patients included in CPRD and the extensive time horizon during which these patients were observed enables us to not only study perinatal outcomes for mother and child but also the risk of developing type 2 diabetes after pregnancy and other postnatal outcomes.
Technical Summary: 
The literature on gestational diabetes uses observational data to study risk factors and outcomes but it has not rigorously analysed the effects interventions such as life style advice or metformin/insulin. On the other hand, clinical trials have investigated the impact of such interventions but are undertaken with a limited time horizon, and only capture pre-specified outcomes. Furthermore, the sample size in clinical trials is often too small to perform adequate subgroup analyses and evaluate heterogeneous treatment effects. In combination with a regression discontinuity (RD) design, CPRD data allows us to exploit multiple advantages. To establish causality, the RD design takes advantage of the glucose threshold rules set by UK clinical guidelines that recommend gestational diabetes interventions. Due to the CPRD’s long time horizon and large sample size, we will not only analyse antenatal and perinatal outcomes but will also evaluate postnatal effects of gestational diabetes interventions in different subgroups of patients. We will evaluate the robustness of results by gradually narrowing down the bandwidth around the treatment threshold and thus only including patients with glucose levels increasingly close to the treatment threshold level. Since gestational diabetes intervention may not only affect mean outcomes but also other distributional features such as the variance, we combine the RD design with a distributional regression approach (GAMLSS, Generalized additive models for location, scale and shape) to evaluate the impact on the whole conditional distribution of the outcomes. The findings of this study are expected to provide novel insights into the effectiveness of gestational diabetes therapy in a real-life setting and can directly inform clinical practice.
Health Outcomes to be Measured: 
Child outcomes: birth weight, gestational age at birth, size for gestational age, macrosomia, mortality, Apgar score, need for respiratory support, still birth/neonatal death, birth trauma (e.g. shoulder dystocia, fracture of humerus or clavicle, brachial plexus injury), neonatal hypoglycaemia, neonatal hyperbilirubinaemia, admission to the neonatology department, childhood obesity, childhood diabetes; childhood asthma, poor mental health. Maternal outcomes: caesarean delivery or other interventions during labour, preeclampsia, plasma glucose and HbA1c levels during pregnancy, post-natal depression, later life outcomes of overweight or obesity, poor mental health, type 2 diabetes and complications (retinopathy, etc.), hypertension, strokes and MIs, renal disease, peripheral vascular disease, all cause mortality, cardiovascular mortality.
Application Number: 
20_000080
Collaborators: 

Till Bärnighausen - Chief Investigator - University of Heidelberg
Maike Hohberg - Corresponding Applicant - University Hospital Heidelberg
Anant Jani - Collaborator - University of Oxford
Anuradhaa Subramanian - Collaborator - University of Birmingham
Christian Bommer - Collaborator - University of Heidelberg
Duy Do - Collaborator - University of Heidelberg
Justine Davies - Collaborator - University of Birmingham
Krishnarajah Nirantharakumar - Collaborator - University of Birmingham
Pascal Geldsetzer - Collaborator - University of Heidelberg
Sebastian Vollmer - Collaborator - Georg-August-Universität Göttingen

Linkages: 
2011 Rural-Urban Classification at LSOA level;CPRD Mother-Baby Link;Patient Level Index of Multiple Deprivation;Pregnancy Register