Which patient groups benefit from thyroid hormone replacement therapy: a regression discontinuity design

Date of Approval
Application Number
20_180
Technical Summary

2% of the UK population – and more than 5% among those aged 60 years and older – suffer from hypothyroidism. Women face a five- to tenfold risk of being affected by the disease. Hypothyroidism is a key risk factor for mortality from cardiovascular disease. Based on evidence from several RCTs, current clinical guidelines (NICE) suggest that patients with a TSH level of 10mU/L or above should receive levothyroxine treatment. However, RCTs are often limited in that they are unable to capture more long-term effects of certain medical treatments. While cohort studies can cover a longer timeline and capture real-life circumstances, they present less consistent evidence on the effectiveness of levothyroxine treatment. In light of this, our study seeks to measure the effect of levothyroxine treatment on short-, mid-, and long-term clinical outcomes based on retrospective routine data provided by CPRD. . Outcomes will include all-cause mortality, severe health events, hospitalizations, and related conditions, including levels of overweight (with BMI cut-offs of BMI>=25, BMI>=30), hypertension (with blood pressure cut-offs of >=140/90 mm Hg, >=160/100 mm Hg) and presence or absence of dementia. To establish causality, we make use of a regression discontinuity (RD) design taking advantage of the TSH cut-off for levothyroxine treatment that is defined in major UK clinical guidelines. Acknowledging that physicians may base their treatment decisions on additional considerations besides clinical guidelines, we use an instrumental variable approach that is robust to non-compliance among some physicians/GPs. Precisely, we will estimate a two-stage approach, using the predicted treatment probability based on the threshold rule as an instrument for actual treatment receipt. Based on this, we then estimate the causal effect of the treatment – that is of levothyroxine therapy – on the clinical outcomes named above (see also section D). We evaluate the robustness of results by gradually narrowing down the bandwidth (e.g. bandwidths that are 50%, 75%, 125%, and 150% of the empirically derived mean squared error-optimal bandwidth) around the treatment threshold and thus only including patients with TSH levels increasingly close to the treatment threshold level. We will estimate “fuzzy” RD models using local linear regression to avoid overfitting data and triangular weights to give more influence to observations close to the threshold. The findings of this study can provide novel insights into the effectiveness of levothyroxine therapy in a real-life setting.

Health Outcomes to be Measured

Primary outcomes: number of severe adverse health events (stroke, heart attack) and all-cause mortality (measured over time horizons of one, five, ten, and fifteen years)

Secondary outcomes: number of all-cause hospitalizations, probability of at least one all-cause hospitalization, falls/fractures, BMI, blood pressure, diagnosis of hypertension, diagnosis of dementia, probability of at least one severe adverse health event (measured over time horizons of one, five, ten, and fifteen years)

Collaborators

Till Bärnighausen - Chief Investigator - University of Heidelberg
Janina Steinert - Corresponding Applicant - Technical University of Munich
Anant Jani - Collaborator - University of Oxford
Justine Davies - Collaborator - University of Birmingham
Pascal Geldsetzer - Collaborator - University of Heidelberg

Linkages

2011 Rural-Urban Classification at LSOA level;HES Admitted Patient Care;ONS Death Registration Data;Patient Level Index of Multiple Deprivation