The impact of waiting time on health gains from surgery: Evidence from a national patient reported outcomes dataset Running title

2Data

We use Patient Reported Outcomes Measures (PROMs) data linked to the Hospital Episode Statistics (HES) database, which contains standardised information on all hospital admissions in the English NHS. The PROMs are patient-level data collected from all providers of NHS-funded care for four large-volume procedures: hip replacement, knee replacement, hernia repair, and varicose veins. The data have been collected since the 1^st April 2009.

Patients are surveyed before and after surgery, using paper-based self-completion questionnaires. The pre--surgery questionnaires collect information on generic health-related quality of life measures (EQ5D-3L (referred to hereafter as EQ-5D), EQ-VAS, self-assessed health) and disease specific measures (the Oxford Hip Score (OHS), the Oxford Knee Score (OKS) and the Aberdeen Varicose Vein Questionnaire (AVVQ)). The pre-surgery questionnaire can be administered to patients on the day of admission by hospital staff, or at any time during the interval between a patient being passed fit for surgery and the intervention taking place ( Department of Health, 2008).

The same patient health information is collected again with the post-surgery questionnaire. In addition, the post-surgery questionnaire collects information on patient satisfaction with the procedure and their perception of the success of the intervention. These questionnaires are sent out by the Health and Social Care Information Centre via a postal survey approximately 6 months after the surgery date in the case of hip replacement and knee replacement, and 3 months after surgery for varicose vein and hernia repair.

The EQ-5D questionnaire asks patients to classify themselves as having one of three levels of health in each of five dimensions of health – mobility, self-care, usual activities (all three scored as no problems/some problems/unable), pain/discomfort and anxiety/depression (both scored as no/moderate/extreme). This results in an EQ-5D health profile for a patient. A preference-based societal utility estimate can also be assigned to the EQ5D profile (Dolan, 1997), known as the EQ-5D index. The PROMs initiative also collects information on the visual analogue scale, the EQ-VAS, which records each patient’s overall assessment of their health on a scale from 100 (best imaginable health) to 0 (worst imaginable health) and patient’s self-assessed health: “In general would you say your health is...”, with five response categories ranging from “Poor” to “Excellent”.

The PROMs database offers a very large sample size of patients containing information on health (frequency and duration of symptoms; previous surgery, disability, pre-existing conditions, post-surgery degree of recovery, disease-specific and generic measures of self-reported health) and socio-demographic characteristics (age, sex, and living arrangements). To measure other pre-existing health conditions, patients are asked “[h]ave you been told by a doctor that you have any of the following?” and are requested to tick all that apply from a list of 12 conditions. The living arrangements question asks patients to report their living arrangements in one of four broad categories: “I live with partner/spouse/family/friends”; “I live alone”; “I live in a nursing home, hospital or other long-term care home”; and “Other”.

The sample we analyse is for patients admitted for surgery between April 1, 2009 and November 1, 2010. This was the latest date for which data from the post-surgery questionnaires were available at the time we requested the data. We only keep observations for which the status of the post-surgery instrument is complete; this leaves us with 60% of hip replacement, 59% of knee replacement, 55% of varicose vein, and 63% of hernia repair records. We also discard a small number of observations with duplicate episode identifiers.

The current waiting times policy guarantee refers to the total period between referral from a general practitioner and receipt of treatment, but we focus on the delay between when the specialist decides that the patient requires treatment and the treatment date. This is commonly called ‘the inpatient wait’. We eliminated just under 1.5% of the sample with excessively long waiting times longer than 30 weeks. These are likely to be coding errors or delays caused by exceptional factors that may be related to the expected gains from treatment (e.g. the patient is not fit for surgery or is undergoing other treatments).

We augment the database by adding two continuous measures of income and education/skills deprivation from the Index of Multiple Deprivation. Socio-economic characteristics may affect waiting times (Laudicella, et al., 2012) and health outcomes. These are measured for 32,482 small geographical areas called lower-level super output areas (LSOAs) and attached to patients on the basis of their area of residence. The income deprivation measure represents the proportion of the population receiving State benefits on the grounds of low income. The education measure is a composite of educational attainment from Key Stage 2 (the stage of learning in the English National Curriculum covering ages 7 to 11 years) of higher education in children and young people and the prevalence of formal educational qualifications in the working-age population.

We analyse samples of 29,303 hip replacement patients, 32,602 knee replacement patients, 9,184 varicose vein patients, and 22,889 hernia repair patients.

3Methods

3.1Model set-up

The timing of the health measures is the most important consideration when modelling the impact of waiting times on patient reported outcome measures. Some of the pre-surgery health variables are time-invariant (e.g. gender, certain health conditions) or time-defined (age, length of symptoms), so we can construct a set of determinants of health as of the date of the consultant’s decision to admit. We study the impact of the length of the period of waiting which starts with the date of the consultant’s decision to admit the patient and ends with the patient’s admission to hospital for surgery.

The analysis needs to account for the fact that hospitals are likely to prioritise their waiting lists according to the health status of patients. We benefit from the rich health information in the PROMs database to control for the potential endogeneity of waiting time. We introduce a broad set of health measures and personal characteristics recorded at the date of the pre-surgery questionnaire. We also introduce a full set of provider dummy variables (or fixed effects) to control for heterogeneity across hospitals in their management of waiting lists and quality of care.

We begin by analysing how time-invariant patient characteristics correlate with waiting times using OLS regression. We then utilise OLS and ordered probit regression models to analyse post-surgery health outcomes as a function of waiting times for surgery. After controlling for a broad set of baseline health characteristics, patient demographic characteristics and provider effects, we test whether longer waits lead to worse post-surgery health outcomes as measured by generic (EQ-5D, EQ-VAS) and condition-specific (OHS, OKS, AVV) measures and self-assessed measures of health and surgery effectiveness.

Assume corresponds to the health status of the -th patient, i=1,.., N, who received surgery at hospital , This variable is measured prior to surgery () and following surgery (). We estimate:

There are unobservable hospital-level effects, . is a vector of individual characteristics measured prior to surgery, is the individual’s waiting time and is the random component. Equation (1) is a value-added approach to estimating the health production function and includes the baseline health outcome which, conditional on the other individual characteristics, is assumed to be a sufficient statistic for the individual’s unobserved health history.

The effect of waiting times on health cannot be signed a priori. Waiting times might affect both pre-surgery health (because health deteriorates during the waiting period) and post-surgery health through a reduced capacity to benefit from the intervention. Hence, the coefficient on waiting times would be negative. Alternatively, patients might be prioritised according to their pre-intervention health to maximise the potential to gain or restore health. In this case, the coefficient on waiting times could be positive.