Objectives To examine how much of the variation between general practices in referral rates and cancer detection rates is attributable to local health services rather than the practices or their populations.

Design Ecological analysis of national data on fast-track referrals for suspected cancer from general practices. Data were analysed at the levels of general practice, primary care organisation (Clinical Commissioning Group) and secondary care provider (Acute Hospital Trust) level. Analysis of variation in detection rate was by multilevel linear and Poisson regression.

Setting 6379 group practices with data relating to more than 50 cancer cases diagnosed over the 5 years from 2013 to 2017.

Outcomes Proportion of observed variation attributable to primary and secondary care organisations in standardised fast-track referral rate and in cancer detection rate before and after adjustment for practice characteristics.

Results Primary care organisation accounted for 21% of the variation between general practices in the standardised fast-track referral rate and 42% of the unadjusted variation in cancer detection rate. After adjusting for standardised fast-track referral rate, primary care organisation accounted for 31% of the variation in cancer detection rate (compared with 18% accounted for by practice characteristics). In areas where a hospital trust was the main provider for multiple primary care organisations, hospital trusts accounted for the majority of the variation attributable to local health services (between 63% and 69%).

Conclusion This is the first large-scale finding that a substantial proportion of the variation between general practitioner practices in referrals is attributable to their local healthcare systems. Efforts to reduce variation need to focus not just on individual practices but on local diagnostic service provision and culture at the interface of primary and secondary care.