What is the impact of regular monitoring using blood tests in people with long term conditions on patient outcomes? Trial emulation using routinely collected primary care data.

Problem

It is generally accepted that people with long term conditions benefit from regular monitoring. However, the evidence base for the optimal monitoring strategies, including which test should be used at what frequency, is weak. Current practice is largely based on expert opinion and local protocols vary, which has led to substantial variation in blood test use within the UK. We aim to investigate whether regular monitoring in people with type 2 diabetes mellitus (T2DM), hypertension, or chronic kidney disease with certain blood tests impacts health outcomes using routinely collected primary care data.

Approach

We are developing analyses to emulate a target trial using primary care electronic health records from Clinical Practice Research Datalink (CPRD) and Hospital Episode Statistics (HES). We are using a sequential trial approach to estimate the effect of regular testing with commonly used blood tests (including liver function tests, renal function tests, and lipid profile) on patient outcomes. We will compare patients who have received regular tests to patients who have not received these tests. The primary outcomes are events that could be prevented with regular monitoring such as unplanned hospital admissions. We will censor patients when they deviate from their assigned strategy, pooling data from the trials to use pooled logistic regression to calculate outcome cumulative incidence and risk difference. Time-varying confounding will be accounted for by applying time updating inverse probability weights.

Findings

We are developing the analysis for liver function tests to monitor people with T2DM and aim to apply these methods to other tests and conditions within the next months. The eligibility criteria for this trial were having a T2DM diagnosis and HbA1c record +/- 30 days of diagnosis between 2004 and 2019, not being pregnant during the study period, and no history of liver disease. 41,778 people were eligible for trial recruitment. Patients were recruited on the date of their first HbA1c test 12 weeks after diagnosis and assigned to the testing strategy compatible with their data on that day. 18,489 people had a monitoring appointment without a liver function test and were assigned to the control group. 29,309 people had a liver function test with their HbA1c test and were assigned to the intervention group. Eighty percent of people in the control group and 56% of the intervention group switched monitoring strategy during follow-up and were censored.

Consequences

We will use these findings to decide whether to recommend regular monitoring with certain blood tests in patients with T2DM, hypertension, or chronic kidney disease. Challenges developing these methods include accounting for residual confounding, high censoring rates, and limitations associated with routine data.