Socioeconomic differences in waiting times for elective surgery: a population-based retrospective study

In Italy the health service is universalistic: each citizen has the right to health care, free of charge or following a co-payment fee, for a wide range of health problems [23] Enrolment on the elective surgery waiting list occurs directly, through the specialist hospital physician or through the patient contacting the chosen hospital under the guidance of the general practitioner. The National Health System is organized on a regional basis: the Regional Health System (RHS) of Piedmont Region is divided into Local Health Authorities (LHA), which provide medium and low-complexity hospital-based healthcare and community care, and Hospital Trusts (HT), providing high-complexity hospital care. Finally, there are a small number of private hospitals, mainly religious, funded by the RHS, that provide medium or low-complexity care.

The HDD is the administrative database that covers all episodes of care for hospital patients (n = 820,000) provided in the hospitals funded by the RHS. The HDD includes information on admission and discharge dates, coexisting medical conditions, patients’ characteristics (sex and age), diagnosis codes (up to 5 digits), surgical procedure codes (up to 5 digits) according to the International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM), the patient’s LHA of hospitalisation and educational level, classified into seven categories. The database is directly managed by the regional authority for information system. The study population consisted of patients undergoing surgery between 2006 and 2008. It is worth noting that since the HDD records discharges some individuals could be registered before 2006.

The surgical procedures analyzed were: coronary artery by-pass (CABG), angioplasty, coronarography, endarterectomy, hip replacement and cholecystectomy. The time on the waiting list was computed as the number of days from registration (the decision to treat or referral) to surgery.

Exclusion criteria for the present study were: 1) missing values from the education variable; 2) providers with very few cases; 3) waiting times higher than the 99^th percentile.

The percentage of missing values from the education variable for the overall sample including all hospitalisations for the 6 selected procedures was 34.6%, 51.3%, 45.9%, 51.3%, 34.2%, 26.2% respectively for CABG, angioplasty, coronarography, endarterectomy, hip replacement and cholecystectomy. In order to reduce this percentage we conducted two record linkage activities: 1) by means of anonymous record linkage between the HDDs over the period under consideration (2006–2008), we were able to reduce the percentage of missing values by attributing the highest level of education reported for the three years of observation, 2) exclusively for residents in Turin, a record linkage based on gender, place and date of birth was conducted between our dataset and the municipal registry, which included information about educational level. After the linkage process, missing values dropped to 12.6% for CABG, 24.6% for angioplasty, 21.8% for coronarography, 24.2% for endarterectomy, 16.1% for hip replacement and 17.4% for cholecystectomy. Table 1 summarises the cases recovered by types of record linkage and the cases excluded from the analysis by reasons for exclusion for each operation.

In order to make the results comparable with international classification, educational level was grouped into three categories: no education to primary education; lower secondary education; upper secondary education and higher education (reference category).

To examine the relation between waiting times and education, we first estimated the cumulative probability of undergoing surgery by waiting time since registration using the product-limit method. The log-rank test was used to compare waiting times across education groups. Cox proportional hazard models [24] without censored data were then estimated to study the relationship between waiting times and educational level taking into account the confounding effect of sex (reference: female), age (reference: > = 75), and registration period (reference: 2006) [25, 26]. Moreover, in order to take into account differences in the severity of clinical conditions, models were adjusted for the Charlson Comorbidities Index (CCI) [27, 28] (reference = no comorbidities).

Lastly, LHAs (reference: LHA - city of Turin) were added in the models to adjust for possible variations in geographical and access management. We estimated models by progressively adding subsets of covariates according to the following pattern: 1) education; 2) age; 3) CCI; 4) LHAs and 5) registration period. Table 2 describes the subset of variables used in the analysis along with their codes and values.

The models were compared using the Akaike information criteria. The assumption of proportional hazards was evaluated:

The instantaneous probability of undergoing surgery at time t, given that a patient has not been operated yet, was expressed as:

where h 0 t is the hazard rate at time t for a reference subject and β _j is the log hazard ratio associated with covariate j. To simplify, assume that there is one covariate x, then the hazard ratio between two subjects with covariate values x ₁ and x ₂ is:

When x 2 = x 1 + 1 the hazard ratio is equal to H R = exp β j and measures the effect of one unit increase in x on the probability of undergoing surgery. If the HR is greater than 1 (β _j >0), the probability of undergoing surgery increases for subjects with covariate value x ₂ compared to subjects with covariate value x ₁, while a HR lower than 1 (β _j <0) indicates a decreased probability of leaving the waiting list. The HR is assumed to be constant over time. To assess this assumption we add a time-varying coefficient γ in the hazard function:

The hazard ratio for a unit increase in the variable x is:

where γ measures the change in the hazard ratio with time, i.e. non-proportionality. If γ>0 (γ<0) than the HR increases (decreases) over time. Testing for non-proportionality of the hazards is equivalent to testing whether γ is significantly different from zero. No ethical approval was required by Italian law 211/2003 which explains why no ethic committee’s permission is needed for this kind of study in Italy (anonymous data from administrative database).

As shown in Table 3, during the study period, most patients who underwent hip replacement (54.7%) and cardiovascular procedures (CABG, 49.1%; angioplasty, 43.2%; coronarography, 46.2%; endarterectomy, 63.1%) were in the group with a lower educational level. Cholecystectomy was balanced across education groups. Most subjects undergoing cardiovascular procedures were male whereas most of those undergoing cholecystectomy and hip replacement were female. At registration, the majority of patients, among those who underwent cardiovascular procedures, were aged 65 and over. The reverse was observed for cholecystectomy, where younger patients underwent more surgery than older ones. The proportion of patients with at least a comorbidity at the time of the registration in the list were particularly high, as expected, among subjects awaiting cardiovascular procedures and, to a lesser extent, among those performing hip replacement and cholecystectomy. As for the geographical location, it is worth noticing that 34.2% of CABGs were performed at the HT of the Piedmont Region; 31.7% of angioplasties and 23.4% of coronarographies at Turin LHA; about two thirds of endarterectomy procedures at the two regional Health Trusts, around 20% of hip replacement surgery at the LHA of Southern Piedmont; and 21.7% of cholecystectomies at Northern Piedmont LHA.

Table 4 summarizes median waiting times of the cohort for the selected elective surgeries by education. Median waiting times for lower educational levels were higher than for higher educational levels for all the selected procedures. Differences were particularly high for endarterectomy and hip replacement. Mean waiting times were systematically greater than median waiting times, indicating that the distribution was highly skewed (data not shown). High geographical heterogeneity was also observed between LHAs.

As measured by the log-rank test, statistically significant differences (p < 0.001) in access to surgery between education groups were observed for CABG, angioplasty, coronarography, endarterectomy and hip replacement, with shorter waiting times for individuals with a higher education level (data not shown).

Table 5 presents the results of fitting for the main effects and Table 6 shows linear-time interaction coefficients with educational level.¹ For all considered procedures except for CABG an inverse and statistically significant gradient between waiting times and educational level was observed: the conditional probabilities of undergoing surgery were lower among people with low and middle education levels than for more highly educated people after adjustment for sex, age, comorbidities, registration period, and LHAs.

The interaction terms between education and follow up time are statistically significant for all procedures except for CABG suggesting that the adjusted hazard ratios associated with education were not constant over time (Table 6). Specifically, hazard ratios higher than 1 show that the difference of probability of undergoing surgery between the two groups decreases with increasing follow up.

In models with statistically significant interactions, both the low and middle level education groups at the start of follow-up show a lower probability of leaving the waiting list compared to the higher level education group with the hazard ratio for the low level education group being slightly lower at the start of follow up compared to the middle level group, statistically significant for all procedures except for CABG. At the start of follow up (Table 5), subjects with a middle level of education registered for angioplasty, coronarography, endarterectomy, hip replacement, cholecystectomy have access to surgery at a rate that is respectively 9.5%, 4.4%, 25.5%, 14.8%, 7.5% lower than for subjects with a higher level of education. The estimated gap between the two groups at the median waiting time decreases to 8.2% for angioplasty, 17.4% for endarterectomy, 8% for hip replacement and 4.5% for cholecystectomy, while for coronarography the estimated gap is constant over time.

Similarly, subjects with a lower education level registered for angioplasty, coronarography, endarterectomy, hip replacement and cholecystectomy have access to surgery at a rate that is respectively 12.0%, 8%, 34%, 23.9%, 16.2% lower than for subjects with a higher education. The estimated difference between lower and higher education in terms of median waiting time decreases to 9.5% for angioplasty, 7.4% for coronarography, 24.9% for endarterectomy, 17.8% for hip replacement, 13.5% for cholecystectomy. The reduction is more pronounced in the lower level education group compared to the middle level education group for angioplasty and endarterectomy but is less pronounced for hip replacement and cholecystectomy. Moreover, the chance of delayed surgery increases for males undergoing angioplasty while it decreases for males awaiting endarterectomy and cholecystectomy. The probability of longer waiting times for surgery are statistically significant for older patients scheduled for CABG while they are shorter for older patients waiting for coronarography, hip replacement and cholecystectomy. Coexisting illnesses extend waiting times for most procedures as well.

The effect of the registration period is also statistically significant for most procedures and tends to decrease over time. More precisely, patients registered after 2006 for hip replacement and cholecystectomy (2008 only) have a higher probability of waiting less than those on the list in 2006 whereas the opposite was observed for those registered for CABG after 2006.

Lastly, the results show a large geographical variation in elective surgery rates for all surgical procedures with differences not spatially structured and often changing over time. Such differences may reflect variations in clinical judgment or resource levels.

In order to study the risk of bias due to missing values we carried out a sensitivity analysis comparing the distribution of the covariates and the outcome in the subgroups made up respectively of subjects with:

The distributions result as being highly homogeneous between them and, in any case, no statistically significant differences were observed; so it is unlikely that the presence of missing values may produce relevant bias in the results.

The study involves the implementation of policies aimed at tackling inequity, acting through revision of enrolment mechanisms in waiting lists in order to reduce inequity in access [30]. In particular, a policy implication for the government is to simplify access to comparative information on waiting times across different providers. Indeed, in Italy a booking system is being implemented: general practitioners will be able to access on-line information of waiting times across providers and book patients directly with the provider with the shortest waiting times. This might reduce inequalities in waiting times by reducing the differences for accessing information systems.

The results of the study also suggest the need for further investigations aimed at exploring the association between waiting times and outcome of care; in fact, if waiting times are a likely determinant of worse quality of care, they may well assume the role of determinants that could amplify the socioeconomic differences in the outcome of medical treatments.

The main limitation of this study is its retrospective, observational nature. Only patients who underwent surgery provide any information, since patients removed from the waiting list without surgery have no chance of contributing their waiting times to this analysis [31].

Another potential source of limitation of the study is the elevated number of missing values for educational level, although the sensitivity analyses show that the risk of bias due to the presence of missing values and possibly due to the record linkage process is very low.

Lastly, our definition of waiting time does not take into account the time between the general practitioner referral and the consultation.