Authors: Katie Saunders and Gary Abel
Large surveys of cancer patients help encourage and inform quality improvements in how health services and care are designed and delivered. These surveys also provide new opportunities for research about inequalities in the processes and outcomes of cancer care.
In England, the Cancer Patient Experience Surveys have been conducted annually since 2010, with high-profile survey findings highlighting high- and low-performing hospitals as well as issues concerning regional variation in care quality. However, researchers need to remember to factor in a number of considerations when interpreting the data and using the findings from such surveys.
One important consideration is understanding the population that the respondents best represent. In the context of cancer research, either incident or prevalent cases of cancer are the most relevant populations that are usually considered. “Incident cases” refers to new cancer diagnoses, while “prevalent cases” means people living with cancer at a particular point in time.
Since patient experience surveys are used to evaluate care quality, recently treated cancer cases are often the most relevant population. For example, people with recent inpatient admissions experience are in a position to best evaluate how a hospital is performing. However, incident cases may be more important when considering the experiences of the diagnosis of each new cancer case, and prevalent cases may be more appropriate to understand the experiences of people living with cancer after their hospital care has finished.
With this in mind, an important factor researchers need to consider when analysing surveys of cancer patients is how respondents were sampled. For example: Have the people being sent the survey had a recent hospital admission, a recent cancer diagnosis, or were they invited to respond through a different way altogether?
More than anything else, the case mix of respondents to surveys of cancer patients is determined largely by the way that the sample is defined, according to recent research by RAND Europe. Specifically, survey respondents recruited on the basis of recent episodes of hospital treatment will have very different profile of diagnoses from both incident cancer cases or prevalent cases. For example, cancers that have a high treatment burden (such as leukaemia, where patients typically undergo multiple hospital admissions for chemotherapy) or require long-term care (such as bladder cancer, where treatment often spans many years) are overrepresented among respondents to the English Cancer Patient Experience Survey.
After identifying the most appropriate population, researchers need to consider at least two more factors—whether there is any variation in early mortality between different groups of patients after inclusion in the survey sampling frame, and survey nonresponse.
Early mortality is higher among older patients, people living in more deprived areas, and among patients with cancer with poorer prognosis (such as lung cancer). These groups may be underrepresented among survey respondents, although short intervals between treatment and survey mail-out should limit most concerns in this context. Differential nonresponse is also an issue, although typically high response rates to cancer patient surveys again limit these concerns.
Respondents to the English Cancer Patient Experience Survey represent a population of recently treated cancer survivors, rather than incident or prevalent cases. For the primary aim of this survey – understanding cancer patients’ experiences of health services – this is the most relevant population of cancer cases. However, it is also worth highlighting that patient surveys often cover a range of aspects of care, from diagnosis through survivorship issues. A survey of recently treated cancer patients may not be representative of all of these potential populations of interest.
Patient survey data can provide unique insights for improving cancer care quality and are a rich resource for research. However, features of survey populations—such as sampling, early mortality, and survey nonresponse—need to be acknowledged and, if necessary, accounted for when analysing and interpreting findings from studies using such data.
Sources:
Abel GA, Saunders CL, Lyratzopoulos G. Post-sampling mortality and non-response patterns in the English Cancer Patient Experience Survey: Implications for epidemiological studies based on surveys of cancer patients. Cancer Epidemiol 2016;41:34-41.
Katie Saunders is an analyst at RAND Europe and Gary Abel is a senior statistician at the University of Exeter.