Limitations

The maximally-adjusted models described throughout this report accounted for differences in the distribution of numerous patient demographics and tumour characteristics between Cancer Alliances that might confound the main association under study. Despite this adjustment, geographic variation in treatment was observed for a number of treatment types. However, some of these geographic differences may be attributable to residual confounding rather than real disparities in clinical practice, such as differential routes to diagnosis (unavailable at the time of analysis), variation in the proportion of patients who died before the primary course of treatment could be started or concluded, or geographic differences in patient frailty.

Additionally, reported analyses do not consider treatments provided in private healthcare settings. Due to the absence of private healthcare data, tumours in private patients will have been incorrectly assigned to the ‘no major surgical resection or chemotherapy’ category. Accordingly, the true proportion of tumours that received treatment will be higher than reported, and with the possibility of differences in private treatment access between Cancer Alliances. If present, such differences will explain some of the variation observed between Cancer Alliances in analyses where tumours assigned to the ‘no major surgical resection or chemotherapy’ category are included. We are not aware of any data sources that allow for a reliable estimate of the degree to which this misclassification occurred.

Charlson comorbidity scores were defined for each tumour by linking to non-ovarian primary cancers in the cancer registry or a pre-defined comorbid medical condition documented within an inpatient setting prior to the diagnosis of ovarian cancer. A list of medical conditions considered and the scoring assigned to each is described in Appendix 8. This derivation is such that the score is dependent on the recording of particular diagnoses during patient admission, and may therefore underestimate the burden of index-relevant comorbidity by missing diagnoses in Appendix 8 that are exclusively documented in outpatient or primary care settings. However, a comparison of Charlson comorbidity indexes derived for a fixed general population cohort of adults aged >20 years found that an index based on secondary care data performed at least as well as one that utilised primary care data for the prediction of case-mix adjusted all-cause mortality⁶.

Despite this, the index may not reflect the true burden of all comorbid disease that may influence clinical decision making. For instance, Appendix 1 shows that 82.8% (n=14,196) of tumours in the cohort were assigned a Charlson comorbidity score of zero, representing tumours in patients without any record of another primary cancer in the cancer registry or a pre-defined comorbid medical condition documented within an inpatient setting prior to the diagnosis of ovarian cancer. That more than four-fifths of tumours in the cohort received a comorbidity score of zero sits at odds with the age profile of the ovarian cancer cohort, and research elsewhere for other tumour sites that demonstrated a broad range of comorbid medical conditions⁷. Nevertheless, the Charlson comorbidity index captures at least some of the variation in the probability of treatment, whereby tumours in patients with higher scores were reported as having lower probabilities of any treatment (Appendix 3), any surgery (Appendix 4) or any chemotherapy (Appendix 5).

Finally, this report described geographic variations in treatment according to the Cancer Alliance of residence at the time of diagnosis. It is possible that some tumours may have received treatments from multiple Cancer Alliances over the course of treatment, which may have differed from the Cancer Alliance at diagnosis. As the Cancer Alliance at treatment can vary over time and according to treatment type, Cancer Alliance at diagnosis was reported for simplicity.