Introduction Stage of disease at presentation of cancer is not routinely recorded in medical records, resulting in large amounts of missing data. We investigated survival trends by ethnicity among 0–29 year olds with cancer while using multiple imputation (MI) to impute missing values of stage.

Method Subjects (n=3534) diagnosed with cancer in Yorkshire (1990–2005) were analysed. Individual imputation models were used to assign stage of disease to four main diagnostic groups; leukaemia, lymphoma, central nervous system (CNS) and other solid tumours. Linear regression was used to impute white blood cell count (WCC) for leukaemia as a proxy for stage and ordinal logistic regression was used to impute for the remaining diagnostic groups. The survival analysis was performed using Cox regression.

Results Missing stage data occurred in 66% of cases for lymphoma, 28% for CNS tumours, and 69% for other solid tumours. WCC was missing for 57% of leukaemia cases. Results of the final analysis showed an increased risk of death for south Asians compared to non-south Asians with leukaemia (HR 1.61; 95% CI 1.01 to 2.55) and lymphoma (HR 2.05; 95% CI 1.09 to 3.87), and a decreased risk for south Asians with other solid tumours (HR 0.50; 95% CI 0.28 to 0.89). There was no significant difference by ethnic group for those with CNS tumours (HR 1.51; 95% CI 0.82 to 2.78).

Conclusion Although stage was missing in two-thirds of cases overall, MI was used to minimise bias and enhance the precision of analyses. This technique therefore offers considerable advantages over other approaches such as complete case analysis or coding missing data as a separate category.