Published data meta-analysis of anthracycline-induced leukaemia

Anthracycline chemotherapy is used to treat several cancer types. It is effective at curing cancer but can cause leukaemia. Randomised data are needed to study the effect of anthracycline on leukaemia risk due to possible confounding from patient selection in observational data.  

Methods have been developed in a published data meta-analysis of anthracycline-cardiotoxicity: Data on doses and events were abstracted from 75 trials and used to model an anthracycline-cardiotoxicity dose-response relationship. Risk ratios for anthracycline doses in modern chemotherapy regimens were then applied to population data to estimate the expected absolute cardiac risks from modern chemotherapy and identify patients at greatest risk of anthracycline-cardiotoxicity. 

The DPhil candidate will lead items 1-4. Items 5,6 will be a collaborative effort between statistician supervisor and the candidate. 

1. Systematic literature review
To summarise the current evidence for anthracycline-related leukaemia.

2. Trial search
Systematic searches will identify eligible trials: randomised trial in cancer patients, treatment including anthracycline in at least one arm and another arm without anthracycline, ≥100 patients per arm, median survival ≥12 months, and information on leukaemias recorded.

3. Data extraction
Information will be extracted on cancer diagnosis, number randomised, year trial opened, protocol cancer treatments, anthracycline dose (total protocol dose, dose per administration) and leukaemia (number and type in each arm)

4. Trial confounding and bias
Additional systematic literature searches of leukaemia risks from non-anthracycline protocol treatments will be conducted to determine risk of confounding by other treatments given in each trial. Risk of bias assessments will be completed using standardised tools.

5. Analysis
For each randomised comparison, the number of leukaemias and number of patients at risk will be summed for both anthracycline and control arms. The increase in leukaemia risk per unit dose will be calculated.

6. Absolute risks
The estimated increase in risk per unit dose will be combined with total-population leukaemia rates to produce age and sex-specific estimates of absolute risk.

A pilot literature search has identified 46 eligible randomised trials, including 45,000 patients and 191 leukaemias. The average risk ratio, anthracycline versus not, was 2.3 (95% confidence interval 1.7-3.1), suggesting that there is sufficient statistical power to produce a dose-response relationship.