Method

Ovary, fallopian tube and primary peritoneal carcinomas (‘ovarian cancers’) were selected into the cohort from the national cancer registry for England⁸ if diagnosed between January 2016 and December 2018. Cases were identified according to the following ICD-10/O-2 codes:

• C56 (malignant neoplasm of ovary)
• C57 (malignant neoplasm of other and unspecified female genital organs)
• C48 (malignant neoplasm of retroperitoneum and peritoneum) - excluding sarcomas: 8693, 8800, 8801, 8802, 8803, 8804, 8805, 8806, 8963, 8990, 8991, 9040, 9041, 9042, 9043, 9044, 8810, 8811-8921, 9120-9373, 9490, 9500, 9530-9582
• D39.1 (neoplasm of uncertain or unknown behaviour of ovary)

Only tumours diagnosed within female patients were included in the cohort.

For the purpose of this report, ‘borderline’ malignancies were excluded as the treatment pathway for this sub-group can look quite different and only a minority of such tumours receive systemic anti-cancer therapy – a key treatment type of interest . Tumours identified via death certificate only are also excluded as these diagnoses would not have been referred for treatment.

Tumours with an ICD-10 site code of C56, C57 or C48 were defined as ‘borderline’ if their morphology code in ICD-O-2 was 8442, 8444, 8451, 8463, 8473, 8472 or 8462.

Tumours with ICD-10 site code D39.1 were defined as ‘borderline’ if their morphology code in ICD-O-2 was 8144, 8260, 8313, 8380, 8381, 8440, 8441, 8460, 8470, 8480, 8481, 9000, 9013, 9014 or 9015.

Treatment dates for each tumour were extracted from multiple data sources in a manner consistent with internal PHE standard operating procedures. Briefly, dates of systemic anti-cancer therapy administrations and major surgical resections were extracted at a tumour level from the cancer registry if they occurred during the primary course of therapy (defined for ovarian cancer as the period between one month prior and up to nine months following diagnosis). Where patients with tumours selected selected into the cohort were known to have not received another primary cancer diagnosis during the 18 months before or after the primary tumour of interest, information from the cancer registry was supplemented with any additional dates available at a patient level from the Systemic Anti-Cancer Therapy (SACT) dataset⁹ and Hospital Episode Statistics (HES) admitted patient care dataset.¹⁰ SACT and HES data are otherwise not used as they are patient-linked datasets where the precise tumour a treatment relates to is not identified. A list of major surgical resections is provided in Appendix 7 and referred to as ‘surgery’ within the body of this report.

Systemic anti-cancer therapies were excluded from the analysis if they pertained exclusively to a supportive regimen, such as the delivery of anti-systemic or analgesic medication for the treatment of cancer symptoms. Throughout the main body of the report, systemic anti-cancer therapy is referred to as ‘chemotherapy’. Radiation therapy was not considered as it is rarely prescribed for ovarian cancers.

Once all relevant treatment dates were extracted, each tumour was assigned to one of the following five groups according to the order in which treatments were delivered:

1. No surgery or chemotherapy
2. Primary surgery with adjuvant chemotherapy (i.e. surgery followed by chemotherapy)
3. Neoadjuvant chemotherapy with interval debulking surgery (i.e. chemotherapy followed by surgery)
4. Chemotherapy but no surgery
5. Primary surgery but no chemotherapy

Based on the above treatment groups, four binary comparison groups were created for use in the treatment analyses described in the main body of this report:

1. Any treatment (groups two to five) versus no treatment (group one)
2. Surgery (groups two, three and five) versus no surgery (groups one and four)
3. Chemotherapy (groups two, three and four) versus no chemotherapy (groups one and five)
4. Primary surgery with adjuvant chemotherapy (group two) versus neoadjuvant chemotherapy with interval debulking surgery (group three)

A number of patient and tumour characteristics were likely to be associated with clinical decision-making concerning how best to treat a particular cancer. Analyses were therefore adjusted for the following variables:

• Tumour morphology (the histological type of the malignancy, e.g. clear cell carcinoma)
• Stage at diagnosis (the size and spread of the tumour)
• Patient age at diagnosis
• Charlson comorbidity index (the burden of comorbid health conditions)
• Area income deprivation

As described later, a sensitivity analysis of any treatment versus no treatment was undertaken that included adjustment for performance status (Appendix 9). This variable was not included in the main set of analyses owing to a high proportion of missing data (57.9%; Appendix 1). Performance status is a measure of a patient’s ability to undertake daily living activities. It is scored according to the adult Eastern Cooperative Oncology Group (ECOG) scale¹¹, which rates physical function from 0 to 4, with a score of 4 indicating complete disability and total confinement to a bed or chair.

Stage at diagnosis was that defined by the cancer registry based on information from multiple sources. Cancer registration staff take FIGO staging data provided by the diagnosing trust via the MDT and review it alongside information from pathology reports and clinical investigations to record the most accurate stage at diagnosis possible. If insufficient staging data were available at the time of analysis, a tumour was defined for this report as ‘stage not recorded’. Tumour stages are numbered from 1 to 4, with a higher value indicating more advanced disease.

Finally, a Charlson comorbidity score was derived for each tumour, drawing on diagnosis data from the cancer registry and the HES admitted patient care dataset. Consistent with a PHE standard operating procedure, comorbid diagnoses were selected if they occurred between three and 27 months prior to the cancer diagnosis of interest. As shown in Appendix 8, a total of 15 medical conditions were considered and assigned values between one and six. Comorbid conditions included myocardial infarction (heart attack), dementia and liver disease. The final index ranged from 0-25, with a higher score indicating a greater burden of comorbid disease. Where a patient had no linkage to HES (as happens for private patients or patients with no inpatient admissions), no score was assigned.

Area income deprivation is reported in quintiles according to the income component of the Index of Multiple Deprivation, which provides a relative measure of income deprivation in the area of a patient’s residence. It was defined for each patient by linking their postcode at the time of diagnosis to a 2011 ONS census Lower Super Output Area (LSOA)¹². This LSOA was then linked to the associated Ministry of Housing, Communities & Local Government 2015 income deprivation quintile¹³.

Geographic variation in treatment was analysed at the Cancer Alliance level according to regions defined in 2018. As shown in Figure 5, these disaggregate England into 19 geographic areas that bring together clinicians and managers from different hospital trusts and other health and social care organisations with the aim of coordinating the diagnosis and treatment of cancer patients in the local area. The Cancer Alliance for each tumour was assigned according to the main residence of the patient on the date of diagnosis.

Descriptive statistics

The statistical significance of differences in the crude distribution of treatment groups by patient demographics and tumour characteristics was estimated using the chi-squared test.

Linear probability models

Each binary treatment comparison group detailed above was added as an outcome variable in separate linear probability models. Covariates were then introduced as explanatory variables in three stages:

• Model 1: Cancer Alliance only
• Model 2: as Model 1, plus adjustment for differences between Cancer Alliances in the distribution of patient age, tumour morphology and tumour stage
• Model 3: as Model 2, plus area income deprivation and Charlson comorbidity score

Linear probability models are equivalent to linear regression with a binary outcome, where standard errors, confidence intervals and p-values are adjusted for heteroskedasticity (residuals that violate the normal distribution assumption due to the outcome for each tumour only taking one of two values). A linear approximation of probabilities when using a binary outcome is considered appropriate when probabilities fall between values of 0.2 and 0.8, representing the range within which a logistic function is largely linear¹⁴. This requirement held for all models under study. Importantly, in contrast to logistic probability models, which are conventionally used in analyses of binary outcome data, linear regression permits the direct comparison of estimates from nested models as new covariates are introduced¹⁵.

Weighted effect coding¹⁶ was applied to each linear probability model such that the sum of all estimates from variable categories reported in each model was equal to zero. Estimates are then interpretable as percentage-point deviations from the sample mean (i.e. from the average probability for the tumour cohort as a whole, weighted according to the number of observations within each category reported by the respective model).

Tumours assigned to a non-specific site morphology were dropped from all linear probability models owing to small numbers (0.4%; n=62; Appendix 1). Tumours with stage 1 disease at diagnosis were also excluded as little regional variation in treatment decisions was expected, given 96.3% (n=3,091) of such tumours were treated with primary surgery only or surgery with adjuvant chemotherapy.

Analyses were undertaken using R version 3.5.3.

Sample sizes

From the original cohort of 17,155 tumours, removal of stage 1 tumours and tumours of non-specific site morphologies left an analytical sample of 13,889 ovarian cancers. In the final main analysis, which concerned the probability of primary surgery with adjuvant chemotherapy versus neoadjuvant chemotherapy with interval debulking surgery, the cohort was restricted to the sub-set of patients who received either of the two treatment combinations (n=6,065).

Funnel plots

For each binary treatment comparison group, Cancer Alliance estimates from Model 1 and Model 3 were extracted and presented on funnel plots. Each point on a funnel plot represents a Cancer Alliance. The standard error is shown on the horizontal axis and provides an indication of the number of tumours diagnosed within the Cancer Alliance. Estimates from Cancer Alliances with a greater number of tumours are more precise, appearing further to the right-hand side of the plot. Each Cancer Alliance is plotted with a radius proportional to the inverse of its estimate’s standard error, providing a quick visual indication as to differences in the size of each plotted Cancer Alliance, as represented by the number of tumours.

The percentage difference in the probability of treatment (overall or a particular combination) is shown on the vertical axis relative to the population average (all tumours combined). A Cancer Alliance with an estimate above the middle line suggests that tumours within the geography were more likely to receive treatment than the population average, with estimates below the line indicating a lower probability.

Two pairs of dashed lines are included on each funnel plot that represent the bounds of statistical confidence around the average value. The inner set of dashed lines represents two standard deviations (SD) from the population average and the outer set represents three SD, being approximately equivalent to 95.0% and 99.7% confidence intervals, respectively. Any observation plotted outside of these dashed lines will have a confidence interval that does not include the average value, and may therefore indicate a systematic deviation in clinical practice that warrants further investigation. However, some random variation in the probability of treatment is expected between regions such that some points will sit outside the dashed lines through chance alone. This should be taken into consideration when interpreting funnel plots (for example, five out of every 100 observations are likely to lie outside the two SD funnel).

Sensitivity analysis

Owing to a sizeable proportion of missing data for patient performance status at diagnosis (57.9%; Appendix 1), this variable was not included in the main analyses reported. Instead, its contribution to a reduction in treatment variation between Cancer Alliances was investigated via a sensitivity analysis. This sensitivity analysis constrained the cohort to the 5,823 tumours documented within patients with a known performance status value at diagnosis, then reported estimates for the ‘any treatment versus no treatment’ model with and without the inclusion of performance status alongside all other a priori covariates. Results from the sensitivity analysis are reported in Appendix 9 and show a strong and expected inverse relationship between performance status and the probability of treatment. Variation between Cancer Alliances from the population average were shifted by between 0.2 and 3.2 percentage points.