Changes in blood sugar and weight accelerated most sharply during the first three months after people with severe mental illness began taking second-generation antipsychotic medicines, a large English study has found. The researchers say the two measures did not always move together, so weight alone may not show every early metabolic change.
The analysis covered nearly 97,000 adults aged 18 to 59 whose GP records showed severe mental illness and prescriptions for olanzapine, quetiapine, risperidone or aripiprazole between 2000 and 2023. The team examined records from up to two years before treatment began and up to two years afterwards, comparing changes in glycated haemoglobin, known as HbA1c, and body weight over time.
HbA1c reflects average blood sugar over recent weeks. In the first 90 days after treatment started, its rate of increase was about 14 times faster than before treatment. Weight rose around nine times faster over the same early period. These are comparisons of the rate of change, not claims that a person’s blood sugar or weight became 14 or nine times higher.
The researchers used an individual-level interrupted time-series analysis. They modelled each person’s recorded measurements before treatment, during the first 90 days and from day 91 through the following two years. This allowed them to compare the direction and pace of change around treatment initiation, rather than simply comparing different groups of people at a single point.
The early pattern varied by medicine and by patient characteristics. Among the four medicines studied, people prescribed olanzapine had the largest early increases: HbA1c rose about 50 times faster and weight around 10 times faster than their respective pre-treatment rates. The study also found faster early changes among women than men. These figures describe differences in observed rates, and do not mean that every person taking a medicine will experience the same change.
Baseline blood sugar was also relevant to the study’s findings. Within two years, type 2 diabetes was recorded in 19 per cent of participants who had prediabetes when treatment began, compared with 3 per cent of those whose blood sugar was in the normal range. Prediabetes means blood sugar is above the usual range but not at the level used to diagnose diabetes. The comparison identifies different outcomes in the two groups; it does not establish that medication alone caused every diagnosis.
Weight and HbA1c did not consistently change in step. A person could show a change in one measure without a corresponding change in the other. The researchers therefore highlighted the importance of considering both measures when monitoring metabolic health. Their findings focus attention on the first three months, when the observed rates of change were greatest, rather than suggesting that later monitoring is unnecessary.
The study examined routine GP records, not a trial in which treatment was assigned by researchers. It can show how recorded measurements changed around the start of treatment, but cannot prove that the medicine caused every change. The researchers also could not account for all factors that might influence weight or blood sugar, including diet, exercise and additional medication. As with other analyses based on health records, the findings depend on what was recorded in clinical practice.
The records covered four oral second-generation medicines and adults aged 18 to 59. Prescriptions recorded in primary care may not mark a person’s first-ever exposure, because treatment can begin in specialist services. Some medicines generally managed in secondary care, including clozapine and long-acting injectable antipsychotics, were not captured in the prescribing data. The results therefore do not describe every antipsychotic or every patient group.
Antipsychotic medicines remain important treatments for serious mental illness, including schizophrenia and bipolar disorder. The study’s findings add detail about when changes linked to metabolic risk may emerge after treatment starts, while its observational design and record-based limitations set boundaries on what can be concluded.
