In this study, we took a mixed-methods approach to data collection and analysis. Our aim was to extract and analyse the existing but unreported epidemiological, health economic and quantitative survey data from within national cancer systems for the first time. Simultaneously, we collected new qualitative data during interviews with patients with cancer in prison and a range of professionals caring for them, working closely with individuals who have lived experience of being in prison or caring for patients with cancer in prison. The work on phases 1 and 2 was conducted simultaneously and we synthesised the findings iteratively as we proceeded to build a comprehensive picture of how cancer occurs and how people are treated and cared for in the English prison population. This final picture then informed the development of priorities and recommendations in phase 3 of the study.
Phase 1: quantitative data
Epidemiological data sources
Comprehensive cancer registration records are made by the National Cancer Registration and Analysis Service (NCRAS) following the receipt of monthly electronic pathology and clinical data from NHS hospitals and death certificate data from the Office for National Statistics (ONS).26
Comparative cancer incidence between the prison and the community population data
From the cancer registry closedown of 2017, we identified all diagnoses of primary invasive cancers, excluding non-melanoma skin cancer [International Statistical Classification of Diseases and Related Health Problems, Tenth Revision (ICD-10) C44], and diagnoses of cervical cancer in situ (ICD-10 D06) diagnosed in persons with a known gender aged between 18 and 120. All cancer diagnoses during the 1998–2017 period were considered for cancer incidence.
We used publicly available prison postcodes, and time periods in which they were active, to identify all cancer diagnoses made for persons registered at a prison postcode at the time of their diagnosis. This pragmatic method of identifying patients in prison had previously been tested in a study of cancer patients in London undertaken by a cancer registry and most recently in national studies of hospital admissions undertaken by the Nuffield Trust. It is not yet possible to identify remand patients.4,27 The total number of cases found was counted as the numerator in subsequent calculations of cancer incidence in prison.
We used ONS population tables to calculate incidence rates for general population, whereas for the prison cohort we used the England prison population (excluding immigration removal centres) mid-year estimates for 1998–2017 that were provided by the Ministry of Justice (MOJ). Data were stratified by year, gender and age groups 18–20, 21–24, 25–29, 30–39, 40–49, 50–59, 60–69, 70–79 and ≥ 80 years. Owing to differences in age bands with the European Standard Population, categories of 18–20 and 21–24 years were created using the rectangular assumption. Age-standardised rates were calculated using the European Standard Population. To compare the incidence rate of cancers diagnosed in prison with cancers diagnosed in the general population, we calculated age-standardised incidence rate ratios for the 20-year period, with 95% confidence intervals (CIs) using Byar’s approximation.28
Length of hospital stay, treatment and survival
To carry out comparative analyses on length of hospital stay, curative treatment receipt and survival, we identified the prison cohort as adult persons with a first primary cancer diagnosed in prison between 2012 and 2017 and excluded the cervical cancer in situ diagnoses, as well as ‘death certificate only’ cancer registrations. We adopted a matched comparative cohort model, where for all persons with cancer diagnosed in prison (n = 883), we randomly selected individuals from the NCRAS cancer registry who were not diagnosed at prison postcodes (i.e. the general population) in a ratio of 1 : 5, matched on 5-year age group at diagnosis, gender, year of diagnosis, cancer site (three-digit ICD-10 code) and disease stage at diagnosis. There were four prison cancer cases where no matching control could be identified, and these were excluded from further analysis. There were 22 cases with fewer than five matching patients.
Thirty-two patients in prison were further excluded from analysis due to vital status data issues, along with their matched cases. The final numbers for analysis of length of stay, treatment and survival are 847 prison cases and 4165 general population cases.
Length of stay
Length of stay was calculated as the number of bed-days for any hospital admission with an admission date that fell within the period of 31 days before and 183 days after the date of diagnosis with a matching cancer diagnosis.
Treatment data
Linkage to the Hospital Episode Statistics (HES) admitted patient care data was carried out using a matching algorithm based on NHS number, date of birth, gender and postcode at diagnosis.
The HES linked data were used to derive information on surgical resections with curative intent. Any surgical procedure recorded between 31 days before and 183 days after diagnosis and with an OPCS-4 (Classification of Interventions and Procedures) code identified as major surgery was used in the analysis. This meant that for one-fifth of tumours no curative surgery could be identified, affecting both prison patients and their matched cases.
Linkage to the radiotherapy data set was carried out using a matching algorithm and ICD-10 code. Any radiotherapy attendance with a start date from 31 days before to 183 days after date of diagnosis and recorded as having curative intent was included in the analysis.
Linkage to the systemic anti-cancer therapy data set was carried out using patients’ NHS number and matching to the ICD-10 code of their tumour. Any systemic anti-cancer therapy with an earliest drug cycle, regimen or administration start date from 31 days before until 183 days after date of diagnosis and recorded as having curative intent was included in analysis.
Confounding factors
Finally, Charlson comorbidity scores were derived from the diagnosis fields of the inpatient HES data from 27 to 3 months prior to diagnosis, using the ICD-10 codes and scoring derived from Quan.29
Ethnicity is recorded in the cancer registry data from different data sources and takes the most frequently reported ethnicity. We used the following main ethnicity groupings in our analyses: white, mixed, combined Asian and Chinese, black, other and missing.
Routes to diagnosis are based on cancer registration, cancer waiting times and HES data, and defined as screen detected (flagged by cancer registry as detected via breast, bowel or cervical screening programme), 2-week wait [urgent general practitioner (GP) referrals with a suspicion of cancer], GP/outpatient referral (routine and urgent referrals where the patient was not referred under the 2-week wait referral route), other outpatient (an elective route starting with an outpatient appointment that is a consultant-to-consultant referral, other referral, self-referral, dental referral or unknown referral), inpatient elective (where no earlier information can be found prior to admission from a waiting list, booked or planned), emergency presentation [an emergency route via accident and emergency (A&E), emergency GP referral, emergency consultant outpatient referral, emergency transfer, emergency admission or attendance], or unknown (no data available from inpatient or outpatient HES or from cancer waiting times or screening).30
We also identified all data on comorbid psychiatric diagnoses in HES data for each population. After careful consideration of data quality, we decided that the diagnosis and reporting of these conditions was likely not to be made in similar ways for each population. We therefore decided not to include these data in our analyses.
Prevalence data
We intended to carry out a new linkage of national MOJ data held within prison clinical systems with cancer registration data held within NCRAS. This would have allowed us to determine not only the incidence of new cancers diagnosed while in prison but also the numbers of previously diagnosed cancers made in the general population before admission to prison. These data would tell us how many people in prison are living with a cancer diagnosis and give us a fuller picture of the burden of cancer within this population compared with the general population. This would, however, have been a labour-intensive exercise for staff working on prison data systems, and we decided that it was not feasible to pursue in the context of the COVID-19 pandemic. This remains, however, an important objective for future research.
Health economic data
We used the same linked patient-level data, patient cohorts and timelines as above, namely patients with an invasive cancer diagnosis identified in NCRAS for the most recent years 2012–7, with prison cases identified based on prison postcode. However, the cohort in this analysis differs slightly, in that patients with inconsistencies in date of death and other recorded dates are not excluded as they are from the survival analysis. HES data included inpatient and outpatient data only. A&E data were not included given the challenges of identifying cancer-specific A&E care for both sets of patients. HES resource use covers 6 months from diagnosis.
Costs
Unit costs were obtained from NHS Reference Costs 2010–1 to 2018–931 and applied to Healthcare Resource Groups (HRGs) reported in the sample, using the most recent costing year for which HRGs are available. Unit costs for inpatient stays were converted to an average cost per bed-day to capture the cost impact of differences in length of stay. These were applied to inpatient bed-days for emergency versus elective (non-emergency) costs. The relevant unit costs were also applied to day cases and regular day or night admissions. For HRGs that ceased to exist prior to 2018–9, costs were inflated to 2018–9.32 Outpatient attendances were costed from 2018 to 2019 unit costs and based on service code and assuming consultant-led service.
In 2009 the costs of escorts and bed-watches were transferred to the NHS, making the NHS responsible for covering the cost of a prison officer escorting a prisoner to hospital attendances, with some exceptions, such as extended lengths of stay. Escort and bed-watches for prison patients were costed as £168 per hospital attendance and £2232 per bed-day, respectively. This is a weighted estimate based on the average time per appointment and the number of prison officers from a 2006 study of escorts and bed-watches33 uplifted to 2018–9 costs using the Services Producer Price Index.34 The costs of escorts and bed-watches may be overestimated given that we do not know if people were released and in the community during the 6 months from diagnosis, or if they were in the approximately 5% of the prison estate who are eligible for release on temporary licence and hence can apply to attend appointments without an escort. As a result, this analysis should be interpreted with caution, representing an estimate only.
Phase 2: patients’ experiences of cancer care
We adopted an exploratory sequential mixed-methods approach comprising two elements: a qualitative exploration of the experiences of diagnosis and care of people with cancer in prison; and an analysis of NCPES data. In this design the quantitative data are secondary and supplement the qualitative data.
Qualitative interview data
Participants were invited to take part in a one-off semistructured interview.
Sample and setting
We employed purposive sampling to recruit a sample reflective of the current prison population (by age and region). Therefore, we aimed to recruit fewer women, as most (96%) people in prison are men.35
In practice, patients who had a diagnosis of cancer were identified and approached by prison healthcare services in six prisons. These services provided patients with a copy of the information sheet and a verbal outline of the study’s aims and what it involved. The patient’s understanding of the aims of the study and methods for data collection were checked before they signed the consent form. Limited information was provided to the interviewees about the interviewers due to time constraints within the prison setting. Their consent form allowed them to assign their gender. Professional participants were recruited using snowball invitations, mailing lists, newsletters and social media. Interviews were conducted either face to face or over the telephone. Professionals provided written consent to participate.
Interviews
Phase 2 comprised qualitative interviews with (1) people with cancer in prison, (2) oncology practitioners, (3) prison healthcare staff and (4) prison custodial staff. All patient interviews were conducted face to face, and professional interviews were conducted either face to face or over the telephone. Professional interviews were conducted by Renske Visser, who was unknown to the participants. Patient interviews were conducted either one to one by Renske Visser, or two to one by Renske Visser and an Expert by Experience (EbE) or fellow researcher Jo Armes. All interviewers were female except one EbE who was male. Both Renske Visser and Jo Armes have PhDs and are experienced in undertaking qualitative interviews. Renske Visser and Jo Armes provided training (6 hours) to the EbE who co-conducted interviews, which included a discussion about their interest in the topic and the potential biases each might bring. Training was augmented by debriefing and mentorship after each interview.
Interview guide and information sheets
For each participant group a separate interview guide and information sheet was developed. Experts from each group were invited to comment on the interview guides to make sure that all relevant topics were covered. Particularly with the patient group, patient and public involvement (PPI) was paramount, and EbE from Revolving Doors Agency (RDA) were involved in designing the information sheet and interview guide to ensure that relevant information was presented in the right tone and at the right literacy level.
Experts by experience
As part of our approach to PPI, three peer researchers with lived experience of prison were involved in various stages of the research. They advised on the practicalities of doing research in a prison environment, commented on information sheets and interview guides, co-conducted interviews and contributed to the analysis of the interviews. This collaboration was instrumental in asking the right questions and setting priorities in our project. The EbE involvement in the study was supported by RDA. For a reflection on this collaboration, see our published article.36
National Cancer Patient Experience Survey data
The NCPES contains around 70 questions covering different aspects of cancer care experience. The survey also allows individuals to assign their own gender. For the analysis included here, NCPES data from 2012 to 2018 were used. People with cancer diagnosed in prison were identified from the national cancer registration based on postcode of residence at diagnosis. Linkage with NCPES is based on patient and tumour identifiers. A matched population approach was adopted to compare the cancer experience of persons diagnosed with cancer while in prison with the experience of persons diagnosed with cancer in the community. Controls were randomly sampled from the NCPES responders not diagnosed in prison and were matched in a 1 : 5 ratio on gender, age group and NCPES data year. Matching was not performed on tumour site, but control sampling was restricted to tumour sites present in the prison cases, excluding males with breast cancer. The final population included 78 cases and 390 controls. Answers to NCPES questions were dichotomised according to preferred answers versus other. Missing answers were excluded. Logistical regression modelling was adjusted for the matching variables and ethnicity and tumour stage to control for potential confounding.
During the initial part of the study, the national process for obtaining additional anonymous qualitative comments made by patients as part of their NCPES responses was put on hold due to a general review of confidentiality. By the time the new application process for these data had moved from the survey provider to NHS England, the three EbE working with us had helped to define the qualitative interview schedule and we had begun using this to collect rich new data. We therefore decided not to pursue an application for the qualitative comments but to compare the quantitative survey results with the themes emerging from our new interviews.
Ethics approval
The National Disease Registration Service (NDRS) has approval from the Confidentiality Advisory Group of the NHS Health Research Authority to carry out surveillance using the data it collects on all cancer patients under Section 251 of the NHS Act 2006. All analyses of national data were undertaken by Jennie Huynh and Margreet Lüchtenborg, who worked within NDRS. Therefore, separate ethics approval was not required for the phase 1 epidemiological, health economic, or phase 2 survey analyses.
The qualitative interview study in phase 2 received favourable research ethics approval from the Health Research Authority (REC 19/LO/1073) and HMPPS (reference 2019-306) on 3 October 2019.
The methods used in the development of recommendations, including the development of a film, co-design groups and the Policy Lab event in phase 3, received approval from King’s College London Research Ethics Committee (reference MRA-20/21-22403) on 23 July 2021.
Modes of analysis/interpretation
Epidemiological data
To assess the trend over time in the male prison population, we calculated the incidence rate ratio per year for the period 1999–2017 with 1998 as the baseline, using Poisson regression adjusted for age. Age was imputed as a continuous variable, providing age-adjusted incidence rate ratio averaged over the 19 years in comparison with baseline.
The Wilcoxon rank-sum test was used to compare the median length of stay between patients diagnosed with cancer in prison and those in the community.
Logistic regression was performed to investigate the likelihood of the patients in prison receiving curative treatment compared with the patients in the general population. Ethnicity, comorbidity and route to diagnosis were considered as confounding factors, but only route to diagnosis was found to improve the model fit with use of the Akaike information criterion. Logistic regression models were adjusted for matching variables (age, gender, disease stage and year of diagnosis) and additional potentially confounding factors (route to diagnosis).
Survival time was calculated from date of diagnosis or date of the start of curative treatment for those undergoing treatment with curative intent until date of death or latest tracing with ONS (February 2019). Survival estimates for the prison cases and population cases were compared using the Kaplan–Meier methodology.
Cox proportional hazards modelling was used to investigate the relative risk of death in the prison population compared with the community population. Cox models were adjusted for matching variables and extended to include stratification for the receipt of any treatment with curative intent to assess the impact of such treatment. The same models were further adjusted for route to diagnosis. The Cox proportional hazards assumptions were assessed by visual inspection of the log–log plots, and scaled Schoenfeld residuals. Assumptions were met for matched variables but not for the receipt of treatment with curative intent, which is why models included stratification for curative treatment.
Health economic data
Descriptive statistics were calculated for the proportion of patients who used each resource use type. Means and standard deviations (SDs) were calculated for attendances and bed-days for patients who used that resource only. Logistic regression was used to calculate the odds ratio (OR) of patients accessing the resource in prison compared with in the community.
As the distribution of costs was unlikely to be normal, we prespecified testing cost differences between prison and comparison patients using general linear models. The suitability of using a log-link was tested for using the link test. The family (normal, Poisson, negative binomial or gamma) was chosen based on Akaike information criteria.37
All ORs and general linear models were adjusted for matching factors (age group, gender, year of diagnosis, cancer type and disease stage) as well as Charlson comorbidity score and ethnicity as a binary of white compared with non-white. Adjusted costs and cost differences were obtained using margins.
Analyses were conducted in Stata v15 (StataCorp LP, College Station, TX, USA).
Sensitivity analysis
The main analysis included patients with unknown NHS numbers. It is possible that incorrect linking was done for these patients and hence they are excluded from a sensitivity analysis. There were also a few HRGs where either we were unable to find the cost or the cost obtained for the HRG was 0. A sensitivity analysis was conducted for HRGs where the cost was unknown or 0 was imputed as the mean cost of all known HRGs > 0.
Qualitative interview data
We conducted 55 interviews and, of these, 53 were recorded and transcribed verbatim by a professional transcription service. On one occasion, the recording device was not permitted in the prison and so the researcher (RV) took notes. Fifty-four interviews were conducted in English and one was conducted in Dutch (by RV). Patient interviews lasted, on average, 60 minutes, while interviews with professionals were shorter, lasting approximately 30 minutes. Interviewers were encouraged to keep written field notes on their experience of conducting the interviews. All patient interviews took place in a separate room in the healthcare facility in each prison. Most professional interviews were undertaken over the telephone.
We used reflexive thematic analysis,38 as this supported an experiential orientation whereby precedence is given to the meaning ascribed by the participant. Initially Renske Visser and Jo Armes inductively coded the interviews. Subsequently our EbE (SW, AX, AB) reviewed transcripts and identified additional codes. Coding continued in NVivo 12 (QSR International, Warrington, UK) and then we reviewed the codes to develop and refine them into themes and subthemes. Analysis of each participant group was undertaken separately, and themes and subthemes were subsequently compared across groups. All analyses were reviewed by the wider research team and our EbE and any necessary refinements were made. The analysis was done both individually by the researcher (RV) and as a wider group with Jo Armes and the EbE from RDA. We held online analysis meetings to discuss common themes. These meetings deepened the analysis as on occasion researchers and EbE interpreted the same data differently. However, after extensive group discussions, consensus was reached on the three core themes. These themes were (1) communication, (2) care and custody and (3) control and choice. As we were specifically interested in cancer diagnosis and management, the themes were mapped to the generic cancer pathway to identify its similarities to or differences from that experienced by cancer patients residing in the community.39
Phase 3: creation of scenarios and public engagement film and development of recommendations
Creation of scenarios and public engagement film
The qualitative interview findings were used to create scenarios depicting the differing perspectives of patients, health professionals in prison and the NHS, and prison professionals. Our aim was to present these to both the co-design and expert groups to enable them to develop their recommendations for improving care based on these and their own experiences. We used a method described by Jones et al.40 to develop vignettes representing the experiences of Black African and Black Caribbean women with breast cancer symptoms in the general population. The steps were as follows:
- Define the target audience to whom they would be presented.
- Collate and analyse the qualitative data, basing the vignettes on interview findings attained through iterative analysis of the transcribed interviews.
- Develop a framework for each vignette – these frameworks were provided by the themes generated in the analysis of interviews.
- Identify and justify quotations – verbatim quotations from the interviews illustrating key themes were incorporated alongside a tabulation of the rationale for their inclusion. This systematic process contributed to their representativeness.
- Create vignettes – these were constructed around themes, associated quotations and ‘alternative viewpoints’. Consideration was paid to length, and presentation was made in the first person. During this process the vignettes were reviewed by the EbE and revisions were made based on their comments.
- Pilot vignettes – this was undertaken to determine whether modifications were needed to the process of introducing and discussing them during the workshops.
- Refine the process of using the vignettes.
- Select narrators – ensuring that each voice sounded authentic to the role being played – and audio-record vignettes.
We partnered with Synergy Theatre Project, an organisation that works towards rehabilitation and resettlement with prisoners, ex-prisoners and young people at risk of offending through theatre and related activities. Synergy Theatre Project identified actors for each part and produced the audio vignettes.
Due to the COVID-19-related lockdown restrictions and work pressures on NHS and prison staff during the pandemic, it was not possible to recruit patient and clinical stakeholders to the co-design workshops that we had planned and designed to develop initial recommendations from the qualitative study findings. Instead, using the vignettes we worked with Synergy Theatre Project and a scriptwriter to develop the audio scenarios directly into a film script that the company then produced as a short 8-minute ‘talking heads’ film entitled Cancer Cells (https://www.surrey.ac.uk/research-projects/how-cancer-care-best-provided-patients-english-prisons-assessing-disease-burden-prison-population#news; accessed November 2024). This was presented during the Policy Lab event as an additional summary of the main qualitative findings and to gain feedback from a wider audience on its strength and tone as a public engagement and dissemination output from the project.
Development of priorities for policy and care recommendations
A Policy Lab is a method designed by the Policy Institute at King’s College London to bring key stakeholders together to consider new data or policy options. This approach recognises that robust evidence is only one ingredient in the development and uptake of new policy and practice based on new evidence. The engagement approach develops aims to provide a forum for open and honest conversations around a policy topic, creating new networks for the synthesis of evidence into an accessible format.41
The event was initially planned as an all-day in-person event, but due to pandemic restrictions and the pressure of work on NHS and prison staff it was designed by the Policy Institute and the research team as an online half-day event and held during March 2022. The event was attended by expert stakeholders from the NHS, MOJ, HMPPS and cancer charities, as well as prison healthcare professionals, EbE, PPI cancer experts and the research team. Invitees were chosen for their personal experience of prison health care and cancer care and for their experience of and potential influence on informing future prison policy planning and development. One week before the event, all attendees received a summary pack of the key epidemiological, health economic, and qualitative and quantitative patient experience findings. In addition, a series of key challenges and barriers identified from the research findings were distilled and outlined. Participants were encouraged to reflect on the evidence and findings emerging from the project, understand the barriers and constraints to change, and develop new ideas and practical approaches to improve outcomes. The key question presented for those at the meeting to answer was ‘What could be the top three improvements to benefit those with cancer in prisons that are achievable in the next three years, given the practical obstacles?’ Participants were enabled to explore how care was organised and delivered and encouraged to think creatively about possible improvements. The practicalities of implementing these changes at a suitable scale were considered, as were the roles of different stakeholders in achieving this.