The labour market consequences of extended waiting times for elective inpatient healthcare: an observational study in England using national data
1Health and International Directorate, Office for National Statistics, Newport, UK
2Strategic Analysis Team, NHS England, London, UK
3Strategy Team, NHS England, London, UK
†Corresponding author: Daniel Ayoubkhani. Address: Office for National Statistics, Government Buildings, Cardiff Road, Newport, UK, NP10 8XG. Tel: +44 1633 455825. Email: daniel.ayoubkhani@ons.gov.ukAbstract
Background
Waiting times for elective healthcare in England have increased since 2015/16, when the constitutional standard was last met. However, little is known about the economic consequences of extended waits. We estimated the labour market shortfall induced by above-target waiting times for elective hospital admissions in England over a nine-year period.
Methods
Cohorts of patients aged 30 to <59 years were identified for 32 treatment specialties using non-emergency inpatient records from NHS hospitals spanning April 2015 to March 2024. Patients were excluded if they had a record within the same treatment specialty from April 2009 to March 2015. Target waiting times were estimated from an exponential distribution with a rate parameter estimated from admissions in 2015/16, assuming patients admitted in each year of the study period remained at their observed percentile of the distribution in that year. Using published trajectories of age-standardised employee pay following elective inpatient treatment, we calculated the additional volume of employment and value of earnings that would be accrued up to five years after the decision to admit, had waiting times remained at target levels.
Results
The study population comprised 5.4 million patients (mean age 47.0 years, 54.3% female), of whom 27.3% were treated in multiple specialties during the study period. The biggest decreases in the maximum wait when moving from observed to target waiting times were seen in clinical haematology (13.6 months), medical oncology (12.5 months) and clinical oncology (10.1 months). Losses to earnings due to above-target waiting times were greatest for trauma and orthopaedics services (£375 million over five years), general surgery (£368 million) and urology (£229 million). Losses to employment were largest for trauma and orthopaedics (9,228 person-years of employment over five years), general surgery (5,570 person-years) and pain management (2,631 person-years). Across all 32 analysed treatment specialties, the total five-year loss was estimated at £1.5-2.3 billion for earnings and 20,502-33,130 person-years for employment.
Conclusions
Extended waiting times for elective inpatient treatment in England are likely to have impacted the labour market. At the macroeconomic level, reducing waits may lead to increased earnings and economic growth, higher tax receipts and reduced welfare expenditure.
Article notes
Competing Interest Statement
The authors have declared no competing interest.
Funding Statement
This study was funded by the UK government Evaluation Accelerator Fund, Phase 4: https://www.gov.uk/government/news/evaluation-accelerator-fund-phase-4-project-summaries
Summary of Updates:
Introduction
The population of England is living longer than ever before, with life expectancy at birth in 2023 increasing by approximately five years over the past four decades and projected to increase by a further four years over the next half-century1. However, these additional years of life are not necessarily being spent in good health2. This fall in healthy life expectancy has been mirrored by increased demand for healthcare services in England, with hospital inpatient activity in 2024/25 growing by over 20% compared with 10 years earlier3,4, outstripping the 8% population growth over the same period5.
The decline in healthy life expectancy and increased demand for healthcare services have coincided with rising levels of economic inactivity; that is, people who are not working nor looking for work. In the final quarter of 2024, 9.3 million working-age people in the UK (over a fifth of the working-age population) were economically inactive, an increase of 8.3% since before the COVID-19 pandemic6. Reasons for labour market inactivity include retirement, studying and caring responsibilities, but the group that has grown most rapidly is those who are not working due to long-term ill-health: up by one-third since pre-pandemic, and now representing nearly 30% of total inactivity among working-age people in the UK6.
There is a growing body of evidence that chronic health conditions can negatively impact employment capacity, potentially with deleterious consequences for the financial wellbeing of individuals, households and the broader macroeconomy. Research has demonstrated that the onset of physical and mental health conditions could reduce annual earnings by approximately £2,000 for the individuals affected and £1,200 for members of their family7.
Incident chronic health conditions, reductions in self-reported general health and onset of disability are also associated with transitions out of employment and into economic inactivity8. The economic impacts of disease may be most pronounced in people with a health condition severe enough to require inpatient care. For example, for patients admitted to hospital in England in 2023 alone, the five-year earnings loss has been estimated at a total of £3.7 billion for cardiovascular diseases, £2.9 billion for cancers and £2.4 billion for musculoskeletal conditions9.
Elective (non-urgent) procedures for people with chronic health conditions that require inpatient hospital care may be effective at improving labour market engagement and increasing earnings10. For example, bariatric surgery has been shown to result in a sustained, long-term increase in the likelihood of paid employment for people with obesity11. However, waiting lists and times for elective procedures such as this have grown over the past decade. In 2004, the UK government outlined its aim of all National Health Service (NHS) patients waiting for consultant-led elective care to be treated within 18 weeks of referral12, and in 2012, it became a requirement for 92% of patients to be treated within 18 weeks. However, this standard was last met in England in 2015/16, with waiting times increasing steadily since then. In 2024/25, just 61% of patients admitted for elective inpatient treatment had waited less than 18 weeks from referral13. Possible reasons for longer waiting times include increased demand for services, and a waitlist backlog caused by service disruption during the COVID-19 pandemic14.
There is substantial evidence on the health consequences of prolonged waits for elective inpatient procedures15–27. Long waiting times may also have health-mediated economic implications, for example through reduced working hours and productivity, or disengagement from the labour market altogether. However, there is a paucity of empirical evidence on the economic impacts of inpatient waiting times28,29. Such evidence could inform government spending decisions in relation to health service delivery, potentially leading to shorter waiting lists and improved patient satisfaction and health outcomes, and ultimately increased labour supply and economic growth; in addition to cutting waiting lists, the latter is a key policy objective of the current UK government30.
This study aimed to quantify the total additional earnings and volume of employment accrued over a five-year horizon if waiting times for a wide range of elective inpatient treatments were returned to their 2015/16 levels. The analysis used electronic health records (EHRs) from NHS hospitals in England, with near-whole population coverage, combined with published estimates of average employee pay and the probability of being a paid employee following elective inpatient treatment.
Methods
Study design and data
This was a retrospective, observational cohort study of working-age patients admitted to NHS hospitals in England for elective treatment, using NHS Hospital Episode Statistics (HES) Admitted Patient Care (APC) data. One study cohort was identified for each of 32 treatment groups (defined by the specialty in which the consultant was working during the period of care). Collectively, these services covered 96% of elective episodes (excluding childbirth-related activity and diagnostic imaging, which are not treatments for health conditions) among working-age patients recorded in HES APC in 2023/24. Low patient counts in each of the treatment specialties covering the remaining 4% of activity precluded analysis for these specialties.
For each of the 32 treatment specialties, patients were included in the study cohort if they: had a decision-to-admit date during the study period 1 April 2015 to 31 March 2024; had been treated by 31 March 2024; had a non-emergency admission for treatment within the specialty of interest (patients could be included in more than one of the 32 study cohorts); were aged 30 to <59 years at the time of admission (this age range was chosen for consistency with the published treatment effects which we used in conjunction with our analytical outputs, described below); and were enumerated at the 2011 Census of England and Wales (required for record linkage within the analytical environment).
For each specialty, patients were excluded from the study cohort if they had a record for this speciality from 1 April 2009 (the earliest available data) to 31 March 2015. This increased the likelihood that patients included in the study population were being treated for a health condition for the first time within the study period. Patients were also excluded if their admission date or waiting time information was missing or invalid, or if their calculated waiting time exceeded two years (deemed to be the result of erroneous data). In instances where a patient had more than one qualifying record within a treatment specialty, the record with the earliest admission date was kept.
An analytical dataset was curated with one row per person per calendar month. The month containing the admission was assigned ‘Month 0’, and patients were followed for up to 12 months before and 60 months after Month 0. Follow-up was censored when patients were aged <30 years or ≥59 years.
Calculating observed and target waiting times
Waiting time was calculated as the difference between patients’ decision-to-admit date and admission date. This is not coherent with the definition used in published NHS waiting time statistics13 (which is based on time from referral rather than decision-to-admit and accounts for mid-pathway clock-stops). However, referral date is not consistently recorded in the HES APC dataset so could not be used for this analysis.
Each patient was assigned a target waiting time, defined as their expected waiting time had they stayed on the same percentile of the observed waiting time distribution (stratified by treatment specialty and financial year of treatment), but the location and shape of the distribution matched that for patients treated in 2015/16 – the latest year when the target of 92% of patients waiting <18 weeks from referral to treatment was last achieved across consultant-led elective healthcare in England13. Candidate statistical distributions for modelling waiting times for admissions in 2015/16 comprised the exponential, gamma, log-normal and Weibull distributions, with the final choice being informed by the Bayesian Information Criterion (BIC).
To align with the estimated treatment effects (see below), which are expressed in terms of calendar months after treatment, the observed and target waiting times were converted from days to calendar months when performing all subsequent calculations.
Observed waiting times being above their target levels during the study period was conceptualised to result in three costs in terms of employment and earnings:
- Accrual of less treatment effects over the five-year period following decision-to-admit
- More time spent in the pre-treatment state accruing disbenefit of illness
- Increased pre-treatment health deterioration means that pay/employment progressed to a lower level following treatment
These components are summarised below and illustrated diagrammatically in Supplementary Appendix A.
Total labour market cost of extended waiting times
For each treatment specialty, the cumulative (across all patients and months) loss to pay/employment due to above-target waiting times over the five-year evaluation period following decision-to-admit was calculated as the sum of: the difference in the cumulative treatment effect under target and observed waiting times; the effect of accruing more disbenefit of disease; the effect of more pre-treatment health deterioration on post-treatment levels.
As the 32 study cohorts are not mutually exclusive, some of the estimated loss in pay/employment due to above-target waiting times for a given treatment specialty may be attributable to patients being treated in other specialities. Thus, there will be some degree of double-counting if estimates of the labour market costs of extended waiting times are simply summed across treatment specialties, giving an upper bound on the estimated total loss due to above-target waits. To provide a more conservative estimate, cohort sizes were re-calculated under the assumption that patients belonging to multiple study cohorts experienced losses in pay/employment due to above-target waiting times for just one specialty in which they were treated: the one with the smallest per-person effect size. The re-calculated cohort sizes were then multiplied by the per-person effect sizes and summed across the 32 treatment specialties, giving a lower bound on the estimated total loss due to above-target waits.
For each treatment specialty, 95% confidence intervals (CIs) were constructed around the point estimates of the change in the cumulative treatment effect under observed and target waiting times by assuming that this quantity follows a normal distribution.
Data preparation was conducted using Sparklyr version 1.9 and statistical analyses were performed using R version 4.4. Further details of the statistical methodology can be found in Supplementary Appendix A.
Results
Description of the study population
The study population comprised 5,363,712 patients (mean age 47.0 years, 54.3% female out of n=5,359,033 with a known sex), derived as per Figure 1. During the study period, 72.7% of patients were treated in one specialty, 20.0% in two specialties and 5.4% in ≥3 specialties. The 32 treatment-specific cohorts ranged from n=11,516 for the anaesthetic service to n=1,615,833 for gastroenterology (Table 1). The study cohorts were generally older and more likely to be female than the general population of England aged 30 to <59 years (mean age 44.0 years, 51.1% female in 202132).
Differences between observed and target waiting times
For modelling waiting times for admissions in 2015/16, no single statistical distribution consistently minimised the BIC across the 32 treatment specialties (Supplementary Table 1). Therefore, the most parsimonious specification (the exponential distribution) was chosen for the main analysis. The gamma distribution was used in a sensitivity analysis as this specification is a generalisation of the exponential distribution; or conversely, the exponential distribution is a special case of the gamma distribution.
The exponential probability density function provided a reasonable representation of the waiting time distribution in 2015/16 for most treatment specialties (Supplementary Figure 1), but with a tendency to over-estimate the proportion of patients with waiting times ≤7 days (this group may include emergency admissions that were not coded as such in the HES dataset), which was not remedied by the more flexible gamma distribution (Supplementary Figure 2).
Across the full study period April 2015 to March 2024, the biggest differences between median observed and target waiting times were for pain management (from 60 to 42 days), cardiac surgery (from 59 to 41 days) and vascular surgery (from 59 to 41 days) (Figure 2). The biggest decreases in the maximum wait when moving from observed to target waiting times were seen in clinical haematology (from 23.4 to 9.8 months), medical oncology (from 23.4 to 10.9 months) and clinical oncology (from 24.0 to 13.9 months).
Labour market impacts of above-target waiting times
Among patients who received a decision to admit and were treated during the nine-year study period, losses to earnings accrued over the five years post-treatment due to above-target waiting times were greatest for the trauma and orthopaedics service (£375 million, 95% CI: £373 million to £376 million), general surgery (£368 million, £367 million to £370 million) and urology (£229 million, £228 million to £230 million) (Table 2). Five-year losses to employment due to above-target waiting times were greatest for the trauma and orthopaedics service (9,228 person-years of employment, 95% CI: 9,201 to 9,256 person-years), general surgery (5,570 person-years, 5,551 to 5,590 person-years) and pain management (2,631 person-years, 2,612 to 2,651 person-years).
Across all 32 analysed treatment specialties and ignoring any double-counting of patients between treatment specialties, the five-year loss due to above-target waiting times summed to £2,281 million (95% CI: £2,264 million to £2,289 million) for earnings and 33,130 person-years (32,887 to 33,374 person-years) for employment. These totals represent upper bounds on the estimated total impact of extended waiting times.
After adjusting the cohort sizes to account for some patients being treated under multiple specialties and multiplying by the estimated per-person effect sizes, the total five-year loss across all 32 specialties was conservatively estimated at £1,470 million (95% CI: £1,460 million to £1,480 million) for earnings and 20,502 person-years (20,363 to 20,641 person-years) for employment. These totals represent lower bounds on the estimated total impact of extended waiting times.
In the sensitivity analysis, very similar aggregate results were obtained when using the gamma distribution rather than the exponential distribution to infer target waiting times, with the total five-year loss conservatively estimated to be £1,489 million (95% CI: £1,478 million to £1,501 million) for earnings and 20,534 person-years (20,381 to 20,687 person-years) for employment. Estimates for individual treatment specialties based on the gamma distribution can be found in Supplementary Table 2.
Discussion
Among all elective inpatients with a decision-to-admit between April 2015 and March 2024 who were treated in NHS hospitals in England, the total five-year loss to employee earnings due to above-target waiting times ranged from £1.5 billion to £2.3 billion. The total five-year loss to payrolled employment ranged from 20,502 to 33,130 person-years. The effect on earnings is likely to be at least partly driven by the effect on employment, in addition to other factors such as reduced working hours and workplace performance.
This study was not intended to be a cost-benefit analysis of policies aimed at reducing waiting times for elective care. However, to contextualise our finding that approximately £2 billion in additional earnings could have been generated if waiting times had remained at the constitutional target, this figure represents less than 0.2% of the total amount spent on healthcare in England over the study period (excluding expenditure directly related to the COVID-19 pandemic)33.
There is substantial evidence on the health consequences of prolonged waits for elective health interventions, with much of the research for inpatients focussing on orthopaedic surgery15–25 as well as prostatectomy15, coronary bypass26 and treatments for varicose veins24,27, hernia24,27 and gallstones27. The results of these studies are mixed, with some but not all finding longer waits to be associated with poorer health and quality-of-life outcomes (possibly reflecting the small samples sizes and specific patient populations studied). Above-target waits for elective treatments have also been found to be associated with increased healthcare utilisation34, demonstrating one potential source of the economic cost of delayed treatment. However, to our knowledge, this is the first study to use individual-level, longitudinal data to explore the labour market consequences of extended waiting times for elective inpatient procedures in England.
One study using administrative data from Norway found increased waits for orthopaedic surgery to be associated with reduced earnings and increased sickness absence28, but it is unclear whether these results can be generalised to other countries and treatments. A study in England attempted to quantify the economic benefit of reducing elective waiting times29, but this was largely based on aggregate rather than record-level data and was in relation to a specific NHS delivery plan35 which is now outdated. Studies in the UK36 and the Netherlands37 have found links between waiting times for mental health treatment and subsequent labour market outcomes, but in relation to specialised, community-based services rather than hospital inpatient settings, which is the focus of the present study.
Strengths and limitations
The main strength of this study is its use of EHRs covering all NHS hospitals in England, thus providing a large sample size (5.4 million patients in total) with whole-population coverage. This meant that results could be produced for individual treatment specialties (collectively covering the vast majority of relevant elective activity) and that study estimates can safely be generalized to the population of interest.
The limitations of this study mainly stem from those of the treatment effects on which it draws, all of which imply that our results are likely to be underestimates of the total economic cost of extended waiting times. The study population only included people aged 30-59 years rather than the broader working-age population (typically defined as those aged 16-64 years in the UK38). The treatment effects only capture the contribution of payrolled employees, not the self-employed (who make up 13% of the UK workforce6). The estimated treatment effects, and thus our estimates of the impact of extended waiting times, do not vary by time waited for treatment. There is evidence of an inverse relationship between treatment effects and waiting time28, perhaps because timely treatment may achieve a bigger effect for progressive health conditions and potentially means less time being detached from the labour market. The treatment effects and our study estimates also do not capture impacts to the economy beyond those experienced by individuals, such as impacts on output/productivity within firms, welfare benefit payments and healthcare utilisation costs.
The HES dataset only includes records for patients treated during the study period; those who were still waiting for treatment at the end of the period are not present on the dataset. Therefore, there will be a degree of right-censoring inherent in our analysis and the cohort sizes will be smaller than the true number of patients with a decision-to-admit during the study period, so the aggregate effect of above-target waiting times will be underestimated.
When estimating patients’ target waiting times, the exponential distribution generally provided a reasonable fit to the waiting time distribution in 2015/16, except for patients who waited a week or less for treatment. This local lack of fit at the lower end of the waiting time distribution is unlikely to have affected the findings of the analysis, as patients with the longest waiting times accrue the most disbenefit of extended waits and therefore contribute most to our estimates.
Confidence intervals around estimates of the impact of extended waiting times do not reflect uncertainty inherent in the input values to the calculations (age-standardised trajectories, modelled treatment and exposure-to-disease effects, and expected waiting times under the target distribution). These are treated as known quantities rather than statistical estimates, so the final confidence intervals are likely to underestimate total uncertainty.
Conclusions
Lengthening waiting times for elective inpatient treatment in England since 2015/16 are likely to have had non-negligible consequences for the labour market, including multi-billion-pound losses to aggregate earnings. This research adds to the growing evidence base on the relationship between population health and the economy, demonstrating that reducing health service waiting times may lead to increased earnings for individuals and their households. At the macroeconomic level, shorter waits may contribute to higher tax receipts, reduced welfare expenditure and, crucially, more economic growth.
The results of this study can be used in economic appraisals of health service and policy interventions aimed at reducing waiting lists for planned inpatient procedures. Indeed, based on this research and estimates of future healthcare demand, the NHS has projected that if waiting times for elective hospital procedures returned to the constitutional standard (92% of patients treated within 18 weeks of referral) by 2030/31, employee earnings in England would increase cumulatively by £2.7 billion over the period and an addition 63,000 person-years of employment would be accrued39.
Further work is needed to investigate the extent to which the labour market outcomes of health interventions are modified by time waited for treatment, which could then be incorporated into refined estimates of the impact of reduced waiting times on aggregate earnings and employment. Also required is research into the labour market consequences of extended waiting times for elective treatments administered outside of inpatient settings, such as outpatient and community services, which were beyond the scope of this study.
Abbreviations
- BIC
- Bayesian Information Criterion
- CI
- Confidence interval
- HER
- Electronic health record
- HES APC
- Hospital Episode Statistics – Admitted Patient Care
- NHS
- National Health Service
- SD
- Standard deviation
Declarations
Ethics approval and consent to participate
Ethical approval for this work was obtained from the National Statistician’s Data Ethics Advisory Committee (NSDEC23(18)) according to the ethical principles outlined here: https://uksa.statisticsauthority.gov.uk/the-authority-board/committees/national-statisticians-advisory-committees-and-panels/national-statisticians-data-ethics-advisory-committee/ethical-principles/
The study also conforms to the World Medical Association’s Declaration of Helsinki – Ethical Principles for Medical Research Involving Human Participants: https://www.wma.net/policies-post/wma-declaration-of-helsinki/
Individual consent to participate in the study was not required for this analysis of administrative data. Sections 45B and 45C of the Statistics and Registration Service Act (SRSA) 2007 provide a power for Crown bodies and other public authorities (respectively) to share the data they hold with the Office for National Statistics (ONS) to produce statistics, without requiring the consent of the individuals to whom the data relate. Although General Data Protection Regulation (GDPR) usually restricts the use of personal data to the purpose for which the data were originally collected, there is special provision within the UK GDPR that permits further use of data for statistical and research purposes (Article 6(1)(e) of the UK GDPR). Further information can be found in the ONS’s health data policy: https://www.ons.gov.uk/aboutus/transparencyandgovernance/datastrategy/datapolicies/collectingandusinghealthdata
Consent for publication
Not applicable.
Data availability
Researchers can apply to access Hospital Episode Statistics data via NHS England’s Data Access Request Service (DARS): https://digital.nhs.uk/services/data-access-request-service-dars
Competing interests
The authors declare no competing interests.
Funding
This study was funded by the UK government’s Evaluation Accelerator Fund, Phase 4: https://www.gov.uk/government/news/evaluation-accelerator-fund-phase-4-project-summaries
Acknowledgements
The authors would like to thank David Goll at Number 10 Data Science, and Chris Worsfold and Henry Foster at the Department of Health and Social Care, for their feedback on the study design. We are also grateful to the following national clinical directors and specialty advisors at NHS England for their valuable insights when interpreting the study results: Dawn Adamson, Mark Cheetham, Jeremy Davis, Jonathan Fuld, Harriet Gordon, Martin Heaton, Tracey Irvine, Lesley Kay, Sue Mann, John McGrath, Adam Millican-Slater, Niranjanan Nirmalananthan, Nick Phillips, Simon Ray, Colin Rees, Alexandra Severns, Doug West.
Supplementary Appendix A: Detailed statistical methodology
Calculating target waiting times
For each treatment specialty (j) in each financial year of the study period (r), we calculated each patient’s percentile (pi) of an exponential distribution with the rate parameter (λ) estimated as the inverse of the mean waiting time :
where nj,r is the number of patients treated in specialty j in year r. For each treatment specialty, the target waiting time distribution was an exponential distribution with a rate parameter estimated as the inverse of the mean waiting time in 2015/16, calculated from the observed HES data:
Each patient’s target waiting time
was then taken from the target distribution (defined by rate parameter λ*), had they remained at the same percentile of the observed waiting time distribution (pi ).
Given that average waiting times were lower in 2015/16 than in other years of the study period, the entire distribution was shifted to the left when moving from observed to target waiting times, with those in the upper tail of the distribution experiencing the biggest changes. By way of example, this is illustrated in Figure A1 for elective Trauma and Orthopaedic Service inpatients, where the median waiting time fell from 57 days (observed for 2015/16 to 2023/24 admissions) to 40 days (target based on 2015/16 admissions).
To align with the exposure and treatment effects described below – which are expressed in terms of month before and after treatment, respectively – the observed and target waiting times were rounded from days to calendar months when performing all subsequent calculations.
Quantifying the labour market cost of reduced waiting times
Observed waiting times being above their target levels during the study period was conceptualised to result in three costs in terms of employment and earnings:
- Accrual of less treatment effects over the five-year period following decision-to-admit
- More time spent in the pre-treatment state accruing disbenefit of illness
- Increased pre-treatment health deterioration means that pay/employment progressed to a lower level following treatment
Cost of extended waiting times, component 1: less treatment effect accrued
We used published estimates of the labour market treatment effects (in terms of employee earnings and person-years of employment) associated with elective inpatient procedures1, the methodology for which has been detailed elsewhere2. In summary, the Hospital Episode Statistics (HES) Admitted Patient Care (APC) dataset was linked to Pay-As-You-Earn (PAYE) tax records from His Majesty’s Revenue and Customs (HMRC) to estimate monthly average employee pay (deflated to 2023 prices) and the monthly probability of paid employment, for months (t) before and after first elective inpatient admission within each treatment specialty. The monthly values (yj,t) were age-standardised to the half-year age distribution in the month of treatment.
A segmented linear regression model was fitted to over the 12 months prior to treatment to infer onset of illness, identified by a breakpoint in the pay/employment trajectory:
where( t − breakpointj)+ = (t − breakpointj) if t > breakpointj and 0 otherwise;
represents the slope of the pre-breakpoint trajectory; and
represents the change in slope after the breakpoint.
The model was used to extrapolate the trajectory from the post-onset, pre-treatment epoch over the post-treatment period. The treatment effect was then estimated as the difference between the observed and model-projected values of age-standardised pay/employment in each month up to five years following treatment: The projected values represent a counterfactual outcome; that is, the expected trajectory of age-standardised pay/employment had individuals not received treatment and instead remained in their ‘pre-treatment’ state. This interrupted time series (ITS) design is illustrated in Figure A2.
We aimed to estimate how much treatment effect (Δj,t) would be accrued within an evaluation period of five years from decision-to-admit under target waiting times, compared with what was actually accrued given the observed waiting times over the study period. We considered two different measures of treatment effect (each relative to the counterfactual of no treatment):
- Additional earnings accrued over the five-year evaluation period
- Additional person-months of employment accrued over the five-year evaluation period
For example, if a patient waited three months for treatment, they accrued treatment effects from Month 0 to Month 57. However, had their waiting time been reduced to one month, they would have accrued treatment effects from Month 0 to Month 59. In general, each patient accrued 60 − wi months of treatment effect given observed waiting times, and months of treatment effect under target waiting times (where
). This notion of accruing additional treatment effect over the five-year period from decision-to-admit had waiting times been at their target levels is illustrated in Figure A3.
Cost of extended waiting times, component 2: more disbenefit of disease accrued
We fitted the same segmented linear regression model as outlined in equation [3] to the published age-standardised pay/employment trajectories1 to estimate the effect of exposure to disease prior to treatment. The pay/employment trajectory from the ‘pre-disease’ epoch (from Month -12 to the breakpoint) was projected over the ‘pre-treatment’ epoch (from the breakpoint to Month -1) to provide a counterfactual outcome: an estimate of individuals’ pay/employment had they remained in the ‘pre-disease’ state, as illustrated in Figure A4.
When waiting times are above their target levels, patients spend more time in the ‘pre-treatment’ epoch, thus they accrue more disbenefit of illness in terms of their pay/employment (∇j,t).
For example, if a patient spent five months in the ‘pre-treatment’ epoch, they accrued five months of disbenefit due to illness (from Month -5 to Month -1). However, had their waiting time been one month shorter, they would have accrued just four months of disbenefit (from Month -5 to Month -2). In general, if patients’ observed waiting times exceed their target waiting times by xi months (i.e. ), they would accrue additional disbenefit of illness from Month -1 to Month -xi, where Month -xi was capped at the breakpoint. This is illustrated in Figure A5.
Cost of extended waiting times, component 3: more pre-treatment health deterioration so pay/employment progresses to a lower level following treatment
When waiting times are above target, patients progress further along the pay/employment trajectory in the ‘pre-treatment’ epoch than they otherwise would have done. Therefore, the post-treatment turning point in the trajectory occurs at a lower level of pay/employment (assuming a downward pre-treatment trajectory), and this difference in level was assumed to be constant and to persist throughout the post-treatment period. As the counterfactual pay/employment trajectory was assumed to be unaffected, the constant that was added to the observed pay/employment trajectory in each month after treatment translated to the same constant added to the treatment effect in each month.
For example, if a patient was treated one month earlier than observed, the change in the pay/employment trajectory that occurred between Month -1 and Month 0 would actually have begun in Month -2. Thus, the difference in level between Month -1 and Month -2 was added as a constant to the observed trajectory (and the treatment effect) from Month 0 to Month 60. In general, if patients’ observed waiting times exceeded their target times by x i months, the constant added to pay/employment in each month post-treatment was equal to the difference in pay/employment between Month -1 and Month [-1-x], where Month [-1-x] was capped at the breakpoint. This is illustrated in Figure A6.
Total labour market losses of extended waiting times
For each treatment specialty, the change in the cumulative treatment effect over the evaluation period following decision-to-admit, plus the disbenefit of more exposure to disease, due to waiting times being above their target levels was calculated as follows:
Boxed Text
[Cumulative treatment effect given target waiting times
−
Cumulative treatment effect given observed waiting times]
+
Effect of accruing more disbenefit of disease
+
Effect of more pre-treatment health deterioration on post-treatment levels
As the 32 study cohorts are not mutually exclusive, some of the estimated loss in pay/employment due to above-target waiting times for a given treatment specialty may be attributable to patients being treated in other specialities. Thus, there will be some degree of double-counting if estimates of the labour market costs of extended waiting times are simply summed across treatment specialties, giving an upper bound on the estimated total loss due to above-target waits:
To provide a more conservative estimate, cohort sizes (nj) were re-derived
under the assumption that patients belonging to multiple study cohorts experienced losses in pay/employment due to above-target waiting times for just one specialty in which they were treated: the one with the smallest per-person effect size. The re-calculated cohort sizes were then multiplied by the per-person effect sizes and summed across the 32 treatment specialties, giving a lower bound on the estimated total loss due to above-target waits:
For each treatment specialty, 95% confidence intervals were constructed around the point estimate of the total labour market loss of extended waiting times by assuming that this quantity follows a normal distribution:
where σ2 is the sample variance of ψ among all patients treated within the treatment specialty. The data were aggregated from person-months to persons when calculating the confidence intervals. Confidence intervals around
and
were derived by using the lower and upper limits of the confidence intervals around
in place of
in equations [6] and [7].