SEARCH

SEARCH BY CITATION

Keywords:

  • waiting time;
  • socioeconomic status;
  • administrative data

ABSTRACT

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. INSTITUTIONAL FEATURES
  5. DATA
  6. METHODOLOGY AND EMPIRICAL ANALYSIS
  7. CONCLUDING REMARKS
  8. REFERENCES

We investigate whether socioeconomic status, measured by income and education, affects waiting time when controls for severity and hospital-specific conditions are included. We also examine which aspects of the hospital supply (attachment to local hospital, traveling time, or choice of hospital) matter most for unequal treatment of different socioeconomic groups. The study uses administrative data from all elective inpatient and outpatient stays in somatic hospitals in Norway. The main results are that we find very little indication of discrimination with regard to income and education when both severity and aspects of hospital supply are controlled for. This result holds for both men and women. Copyright © 2013 John Wiley & Sons, Ltd.

INTRODUCTION

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. INSTITUTIONAL FEATURES
  5. DATA
  6. METHODOLOGY AND EMPIRICAL ANALYSIS
  7. CONCLUDING REMARKS
  8. REFERENCES

Waiting times for elective treatment are a distinguishing feature of a public healthcare sector. The basic reason is that access to care is free, or copayments are low, so demand will exceed supply in terms of short-run treatment capacity. Waiting times impose a deadweight loss because they are costly to patients and entail few benefits for the providers. Still, waiting times might be preferred over rationing mechanisms based on copayments. One important reason for this is that rationing by waiting times is supposed to be independent of socioeconomic status and thus considered more equitable than rationing by copayments. This argument is significant in many national healthcare systems.

Whether rationing by waiting time is independent of socioeconomic status is an empirical question. The purpose of this paper is to contribute to this issue in three ways. First, we investigate whether socioeconomic status, measured by income and education, affects waiting time when controls for severity of illness are included. Second, we investigate whether any such inequalities are due to geographic variation in the supply of hospital services or by unequal treatment of different socioeconomic groups. Third, we examine which aspects matter most for unequal treatment of different socioeconomic groups.1 The aspects we are considering are traveling distance from the patient's residence to the closest hospital, patients' choice of treating hospital, and whether treatment is given at a local hospital or at a university hospital.

We are able to investigate these issues because we have access to a rich data set from a country with the appropriate institutional settings. The data we use are patient-level administrative data from the Norwegian Patient Register (NPR). This data set includes all patients treated by Norwegian hospitals. We focus on elective care patients. Hence, we exclude acute care patients, that is, patients that are directly brought to hospital emergency rooms. In Norway, patients are required to see their primary care physician (general practitioner) to obtain a referral to specialized care. Patients referred from their general practitioners will show up with a waiting time of at least 1 day. Socioeconomic status is measured by small-area-level education or income.

When we investigate the sample of all elective patients, we find evidence of discrimination with regards to both income and education when only controls for severity of illness are included. These effects are quite large, as men with tertiary education wait about 14.5% shorter than men with only primary education. For women, evidence of discrimination is found with regards to income as women in the lowest income quintile wait 11% longer than women within the highest income quintile. However, when controls for hospital-specific factors such as attachment to local hospital, traveling time, or choice of hospital are included, we find very little evidence of discrimination. Specifically, we find that local hospital explains the income gradient and that educational differences in waiting time are mainly due to variation between educational groups in local hospital and travel distance for men and travel distance for women. Hence, these latter factors are the ones that matter most for unequal treatment of different socioeconomic groups.

We are not the only paper that investigates the relationship between waiting times and socioeconomic status. Siciliani and Verzulli (2009) analyzed whether patients with higher socioeconomic status measured by educational attainment have lower waiting times for specialist consultation and nonemergency surgery using data from the Survey of Health, Aging, and Retirement in Europe. The main result is that higher socioeconomic status contributes to a significant lower waiting time in many European countries with a National Health Service type of health system. One limitation with this study is that it makes use of survey data. The sample size is relatively small, and waiting time information is self-reported. In addition, the Survey of Health, Aging, and Retirement in Europe data do not contain any information on supply-side factors, which excludes the possibility to control for any such differences in the empirical analysis.

Cooper et al. (2009) and Laudicella et al. (2012) circumvented problems related to survey data and used administrative data to investigate whether socioeconomic status affects waiting time. Patient-level data from administrative databases are linked with small-area socioeconomic variables. Cooper et al. (2009) investigated changes in waiting times for key elective procedures (hip replacement, knee replacement, and cataract repair) in the English National Health Service between 1997 and 2007 and analyzed the distribution of those changes between socioeconomic groups. A patient's socioeconomic status is measured by the Carstairs index of deprivation, and the data used are from the Hospital Episode Statistics database in England.2 The main conclusions are that waiting times went down, and the variation in waiting times across socioeconomic groups was reduced. Laudicella et al. (2012) did also make use of data from the Hospital Episode Statistics. They investigated whether waiting time for inpatient hip replacement differs according to socioeconomic status measured by small-area-level income and skill deprivation from the indices of Multiple Deprivation 2004. Because small-area-level data are collected during the census in 2001, the analysis focuses on the year 2001/2002. The study does also include controls for severity (the type and number of diagnosis) and supply (hospital-level fixed effects). The authors find evidence of inequality in waiting times favoring more educated individuals and, to a lesser extent, richer individuals: Compared with patients with least skill deprivation, patients in the second quintile wait about 22 days longer (9%), and patients in the third to fifth quintiles wait about 32 days longer (13%).3

Our study differs from these studies in various ways. Most notably, it differs from that of Siciliani and Verzulli (2009) because we use administrative data instead of survey data. With respect to the two latter studies our approach is different because it presents data and results from a country other than England. Like Laudicella et al. (2012), we include controls for severity and controls for hospital-specific conditions. Our controls for hospital-specific conditions include controls for local hospital, distance to hospital, and whether treatment is given at the local hospital or at a university hospital.

The paper is organized as follows. In Section 2, we give a short description of the Norwegian specialized healthcare sector. Section 3 presents the data and the methodology, whereas Section 4 contains the empirical analysis. Concluding remarks are given in Section 5.

INSTITUTIONAL FEATURES

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. INSTITUTIONAL FEATURES
  5. DATA
  6. METHODOLOGY AND EMPIRICAL ANALYSIS
  7. CONCLUDING REMARKS
  8. REFERENCES

The Norwegian specialized healthcare sector is predominantly publicly owned and organized as state-owned enterprises within five (north, mid, west, south, and east) regional health authorities (RHAs). 4 The RHAs have the responsibility for providing specialist health care to all patients within the region. The RHAs receive an annual budget from the Norwegian Government, based on a weighted capitation formula. In addition, the RHAs receive an activity-based grant whose size is proportional to the number and composition of hospital treatments. The activity-based component is about 40% of the somatic budget.

Provision of specialist health care is organized through health enterprises (hospitals) owned and governed by the RHAs. These organizations can also contract with private suppliers for providing treatment. This outsourcing is in effect quite small compared with the overall treatment activity and confined to a few diagnoses.

Some hospitals, denoted local hospitals, are responsible for providing hospital treatment to the population within their catchment areas. A patient may have different local hospitals for different types of treatment. Patients are free to choose a hospital at the national level but cannot choose to receive treatment at a university hospital. The university hospitals have two functions: they are local hospitals for the population in their catchment areas, and they provide advanced hospital treatment for the health regions (or the country for rare diseases). We will classify each patient episode in one of three categories: (i) treatment at the patient's local hospital; (ii) treatment at another hospital except university hospitals; and (iii) treatment at a university hospital that is not the patient's local hospital. Treatments at hospitals in category (ii) are usually due to patient choice, whereas the decision to offer treatment at a category (iii) hospital is not made by the patient.5 In the period covered by our study, few patients received treatment outside the catchment area of their local hospital, Vrangbæk et al. (2007). There are substantial travel distances in Norway, and reluctance to travel is large (Monstad, 2007).

DATA

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. INSTITUTIONAL FEATURES
  5. DATA
  6. METHODOLOGY AND EMPIRICAL ANALYSIS
  7. CONCLUDING REMARKS
  8. REFERENCES

The empirical analysis makes use of data merged from three data sets. The first data set is the NPR for the period 2004–2005. This individual-level register contains information about waiting time and patient characteristics such as age, gender, place of residence (municipality or part of city), main and secondary diagnoses, and surgical procedure codes for all elective inpatient and outpatient treatment in somatic hospitals. The waiting time is measured from referral until the patient meets with a specialist from the hospital. This indicates starts of treatment, even though further diagnosing of the patient may occur.

The second data set is compounded from the tax and education registers of Statistics Norway and covers the population aged 25–66 years.6 Because the NPR (at least so far) does not have a unique personal identifier, information about socioeconomic status cannot be linked at the individual level. However, because the register has information about each hospital stay according to gender, year of birth, and resident municipality, patients can be uniquely assigned to population cells that combine gender, age, and municipality. For each population cell, Statistic Norway has computed a set of variables that describe the income and educational levels of the cell population in 2004. The income variables are mean cell income,7 within-cell standard deviation of income, and within-cell difference between the 90th and 10th percentiles of income. The education variables are the share of the persons in the cell with no more than compulsory schooling, the share of persons that had completed secondary education but no tertiary education, and the share of persons in the cell that had completed at least one year of tertiary education (college/university). We have income and education variables for 31,165 population cells.

The third data set measures the distance from a municipality to the closest municipality with a hospital. Distance is measured in travel time by car from one municipality center to another.

Not all patient episodes are included in the analysis. First, we focus on patients with a date of referral during the first 8 months of 2004 that received treatment during 2004 or in 2005.8,9 During this period, 1,034,788 hospital episodes took place. Second, we drop prenatal care visits (73,345 observations dropped). Third, we drop 14,174 observations from one hospital because 98% of the patient episodes are missing waiting time information. Finally, to avoid serial hospital admissions, we only include the first hospital stay for each patient (335,855 observations are dropped). Because NPR does not have a unique personal identifier, we assume that all hospital episodes where a patient is given the same date of referral and has the same sex, age, main diagnosis, and place of residence (municipality or part of city) refers to one and only one patient.10 After excluding these observations, we are left with a total of 611,414 patient episodes at 73 different hospitals.

METHODOLOGY AND EMPIRICAL ANALYSIS

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. INSTITUTIONAL FEATURES
  5. DATA
  6. METHODOLOGY AND EMPIRICAL ANALYSIS
  7. CONCLUDING REMARKS
  8. REFERENCES

Figure 1 shows the distribution of waiting time, and Table 1 presents summary statistics. The density of waiting time is monotonically decreasing with a long right tail. Men experience shorter waiting time than women; average waiting time for women is about 5 days longer than for men. Average waiting time is more than two times higher than median waiting time (79 versus 34 days for men; 84 versus 38 days for women). Sixty-one percent wait less than 2 months, whereas 3.5% wait more than 1 year. The mean number of diagnoses (main and secondary) is 1.2 for both sexes, and the number of procedures is 2.51 for men and 2.55 for women.

image

Figure 1. The distribution of waiting time

Download figure to PowerPoint

Table 1. Variable description and summary statistics
 MeanSDMedianMaxMin
  1. 611,414 patient episodes. Income and educational achievement: 31,165 cell data points

Male patients (265,315 patient episodes, 15,336 cell data points)
Waiting timeDays78.800113.90034.000719.0000.000
Age 47.80011.60049.00066.00025.000
DiagnosesNumber of diagnoses (main + secondary)1.1900.6001.0008.0001.000
ProceduresNumber of operation procedures2.5101.1702.00010.0000.000
Mean incomeAverage earnings less capital income of cell population (2004, in NOK105)2.9201.0003.03055.0000.000
SD incomeStandard deviation of earnings less capital income of cell population (2004, in NOK105)1.8502.2001.65021.5000.000
90–10 income differenceDifference between 90th and 10th percentiles of earnings less capital income of cell population (2004, in NOK105)4.1301.4104.00028.6000.000
Secondary educationShare of cell population with upper secondary but not tertiary education0.6460.1420.6501.0000.000
Tertiary educationShare of cell population with at least 1 year of college/university education (2004)0.1940.1070.1881.0000.000
SizeCell population size79.000230.40033.0005487.0001.000
Female patients (346,099 patient episodes, 15,829 cell data points)
Waiting timeDays84.300114.90038.000729.0000.000
Age 45.60012.00046.00066.00025.000
DiagnosesNumber of diagnoses (main + secondary)1.2000.6401.0008.0000.000
ProceduresNumber of operation procedures2.5501.0802.00010.0000.000
Mean incomeAverage earnings less capital income of cell population (2004, in NOK105)1.8600.5801.9704.7900.000
SD incomeStandard deviation of earnings less capital income of cell population (2004, in NOK105)1.1900.3401.17012.9000.000
90–10 income differenceDifference between 90th and 10th percentiles of earnings less capital income of cell population (2004, in NOK105)2.9700.7803.03013.1000.000
Secondary educationShare of cell population with upper secondary but not tertiary education0.5910.1340.5921.0000.000
Tertiary educationShare of cell population with at least 1 year of college/university education (2004)0.2520.1400.2401.0000.000
SizeCell population size75.100223.50030.0005643.0001.000

Turning to the socioeconomic data, we see that there are large income differences within population cells as well as between population cells. Across population cells, the unweighted average of mean income is NOK292,000 ($50,000) for men and NOK185,000 ($30,000) for women. Mean cell income varies from NOK5.5m to 0 for men and from NOK480,000 to 0 for women. Within cells, the standard deviation of income is, on average, two thirds of mean income, whereas the difference between the 90th and 10th percentiles is, on average, larger than mean income. There are also large variations in population cell size, from more than 5000 persons in the most populous cells to one person in the smallest cells.11 In Section 4.1, we examine the robustness of our results when population cells with large income dispersion are excluded from the analysis.

To characterize the education level of the cell population, we compute the share of persons within the population cell that had completed secondary education (but not tertiary education) and the share of persons that had completed at least 1 year of tertiary education. The population share with tertiary education is higher among women (the average between cell shares with tertiary education is 25% for women versus 19% for men), whereas the share with compulsory schooling only is the same for men and women (16%).

To check whether the relationship between mean income and waiting time is nonlinear, we include dummy variables for sex-specific and age-specific income quintiles. For each birth year cohort, male (female) patients are placed in one of five income categories depending on mean income in the cell that he (she) belongs to. The income categories comprise 20% of the national male (female) population in every birth year cohort. The highest income quintile consists of the cells that have the highest mean income in the birth year cohort, the second highest income quintile consists of the cells that have the second highest income in the birth year cohort, and so on.

From the NPR data set, we have detailed information about patients' medical conditions. The data set contains information about patients' main diagnosis, the number of secondary diagnoses (up to seven), and if there are any surgical procedure codes (up to 10). To identify the relationship between waiting time and socioeconomic status, we include fixed effects for medical conditions to analyze variations in waiting times for patients with the same medical condition.

In the analyses, we include three alternative specifications of medical conditions. The idea is to check whether the effects of socioeconomic status are sensitive to how well the seriousness of illness is controlled for. The first specification includes fixed effects for International Statistical Classification of Diseases and Related Health Problems (ICD)-10 main diagnosis.12 There are 6343 (5743) different main diagnoses for female (male) patients in the sample. The second specification includes fixed effects for combinations of main and secondary diagnoses. This way of specifying medical conditions gives 33,045 (26,586) different combinations for women (men). The third alternative includes fixed effects for all combinations of main diagnosis, secondary diagnoses, and surgical procedure codes. This specification of medical conditions gives rise to 92,807 (77,623) different groups of female (male) patients.13

In addition to controlling for severity (medical condition) we include controls for several hospital-specific conditions; the purpose is to examine how the socioeconomic gradient in waiting time is affected by hospital supply characteristics and patient choice of hospital. First we include fixed effects for local hospitals. There exists no official classification of local hospital catchment areas according to diagnosis. We choose to define a local hospital to be the hospital that has the highest number of treatments for a given municipality, sex, and main diagnosis (first letter in the ICD-10 code).14

Second, we include a variable describing the distance to the closest hospital from the municipality center. This variable is interacted with age (10-year cuts) because the effect of travel distance might vary with age. Our rationales for including this variable are as follows: (i) people living closer to the hospital might fill open slots on short notice and hence obtain lower waiting time and (ii) people living close to the hospital might also show up at the hospital without a referral but still receive elective treatment, for instance, by seeing a specialist during a visit to an emergency primary health care center. Third, we include a dummy variable for patient choice that is turned on if the treating hospital is a category (ii) hospital, that is, neither local hospital for the patient nor university hospital. Finally, we include dummy variables for the university hospitals. These dummy variables shall capture referrals to university hospitals by primary care physicians or lower-level hospitals.

To analyze how socioeconomic status affects waiting time, we first estimate the following ordinary least square (OLS) model.

  • display math

where wti is the log of the number of days (plus 1) between the days of referral and admission, AGE1 is an age vector, MC is a vector of medical conditions, HC is a vector describing the hospital supply, AGE2 is a vector of 10-year age cuts, D is the traveling time to the closest hospital, CHOICE is a dummy variable indicating whether treatment took place at the local hospital, SES is the patient's socioeconomic status (income or education level), and εi is an error term. The scalars δ0 and δ5 and the vectors inline imageinline image and inline image are parameters to be estimated.

We first consider the effects of mean cell income on waiting time. Persons in cells with the highest mean income are the reference category. The results are presented in Table 2. We present the results for men and women separately.

Table 2. Effect of income on log (1 + waiting time)
 (1)(2)(3)(4)(5)(6)(7)
  1. t-statistics (absolute values) clustered at cell level reported in parentheses.

Male patients
Top income quintileReference category
4th income quintile0.176 (3.510)0.088 (2.860)0.082 (2.640)0.063 (2.660)−0.023 (1.270)−0.021 (1.220)−0.021 (1.170)
3rd income quintile−0.020 (0.340)0.043 (1.420)0.048 (2.240)0.990 (4.300)−0.026 (1.370)−0.024 (1.240)−0.021 (1.100)
2nd income quintile0.077 (1.690)0.084 (3.160)0.074 (2.750)0.156 (6.700)−0.040 (1.990)−0.040 (1.960)−0.036 (1.800)
Lowest income quintile0.220 (5.520)0.174 (7.510)0.171 (7.210)0.235 (11.030)−0.009 (0.420)−0.029 (1.400)−0.026 (1.270)
R20.0070.3170.3880.4800.6470.6470.648
Female patients
Top income quintileReference category
4th income quintile−0.311 (6.500)−0.193 (5.320)−0.180 (5.140)−0.076 (2.740)−0.039 (1.680)−0.036 (1.600)−0.035 (1.690)
3rd income quintile−0.160 (3.660)−0.113 (3.520)−0.095 (2.960)−0.006 (0.230)−0.009 (−0.380)−0.022 (−0.940)−0.020 (0.890)
2nd income quintile−0.057 (1.470)−0.042 (1.400)−0.028 (0.940)0.061 (2.430)−0.0001 (0.004)−0.025 (1.060)−0.022 (0.980)
Lowest income quintile0.017 (0.460)0.007 (0.250)0.027 (0.970)0.110 (4.510)0.022 (1.380)0.010 (0.440)−0.008 (0.360)
R20.0090.2840.3610.5920.6070.6070.609
Fixed effects
Birth yearxxxxxxx
Main diagnosis x     
Main diagnosis × secondary diagnoses  x    
Main diagnosis × secondary diagnoses × procedures   xxxx
Local hospital    xxx
Travel distance × age cuts     xx
Choice variables
Other hospital than local hospital      x
University hospital fixed effects      x

First, we present the result when only fixed effects for birth year are included. The next three columns include the three alternative fixed effects for medical specifications. In regression (2), we control for main diagnosis, and regression (3) controls for combinations of main and secondary diagnoses, whereas regression (4) in addition controls for procedures.

Income has a strong and significant negative effect on waiting time for men. Notice that the results are not much affected by whether and how we control for medical condition. But the effect of income becomes more linear the more extensive the control for medical condition is; the relationship between income quintile and waiting time is almost linear in regression (4). Patients in the lowest income quintile wait about 24% longer than patients in the reference category. For women, the results are less uniform. Now, it matters how we control for medical condition. Without such control, the relationship between income and waiting time is U-shaped, regression (1); the highest and lowest income quintiles wait longest. But when maximal medical controls are included, regression (4), we see a clear prorich bias. Women in the lowest income quintile wait about 11% longer than women in the highest and third highest quintile and about 19% longer than women in the second highest income quintile.

In regression (5), we include dummy variables for local hospitals. For both sexes, local hospital explains most of the income gradient. Only men in the second lowest income quintile have a waiting time that is significantly different from the waiting time of the reference group (4.0% lower). For women, those in the second highest income quintile have lower waiting time (3.9%).

For each local hospital, we compute average age-adjusted, diagnosis-adjusted, and procedure-adjusted waiting time for the patients in the hospital's catchment area and the share of these patients in the five cell income categories. For male (female) patients, the weighted correlations15 across local hospitals between average adjusted waiting time and patient income shares are −0.18 (0.04) for the top income quintile and 0.21 (0.10) for the lowest income quintile. Because hospitals located in low-income regions have longer waiting time than hospitals in high-income and middle-income regions, controlling for local hospital makes the income gradient flatter

We next consider the impact of traveling distance on the income gradient. The income of patients decreases in travel distance, and waiting time increases in travel distance.16 Including traveling distance should therefore weaken the association between income and waiting time. A comparison of regressions (5) and (6) shows that this is the case, but the impact on the income gradient is small.

Finally, we consider how the income gradient is affected by patient choices and treatments at university hospitals. Table 3 shows average adjusted waiting time and the share of patients in the cell income quintile for our three hospital categories. Average adjusted waiting time is lower for patients that choose another hospital: 72 versus 77 days for men and 78 versus 81 days for women. Waiting time for treatment at a university hospital is somewhat longer compared with that at a local hospital, perhaps reflecting a trade-off between waiting time and quality of treatment.

Table 3. Average adjusted waiting time and the share of patients in the cell income quintiles and educational levels for the hospital categories
 MenWomen
Local hospitalsOther hospitalsUniversity hospitalsLocal hospitalsOther hospitalsUniversity hospitals
  1. The mean waiting times are significantly different from each other. (The results are available upon request.)

Mean waiting time (days)77.0072.0083.0081.0078.0086.00
Share in highest cell income quintile0.130.140.180.120.200.33
Share in second highest cell income quintile0.120.170.130.230.160.17
Share in mid cell income quintile0.170.130.110.230.170.16
Share in second lowest cell income quintile0.180.150.150.210.230.16
Share in lowest cell income quintile0.410.400.420.200.240.18
Share in tertiary education0.260.280.280.310.290.34
Share in secondary education0.590.570.580.540.550.52
Share in primary education0.150.150.140.140.160.13

A comparison of regressions (6) and (7) shows that patient choices and treatments at university hospitals have limited impact on the income gradient. This result reflects that there are no systematic income differences between patients treated in their local hospitals and patients choosing other hospitals (Table 3).17 University hospitals have a larger share of patients in the highest income quintile than the other hospital categories, but there are too few treatments at university hospitals to have an impact on the estimated income gradient of waiting time.

We then turn to the effects of educational achievement on waiting time. The results are presented in Table 4.

Table 4. Effect of educational achievement on log (1 + waiting time)
 (1)(2)(3)(4)(5)(6)(7)
  1. t-statistics (absolute values) clustered at cell level reported in parentheses.

Male patients
Secondary education0.408 (2.670)0.365 (3.840)0.383 (3.880)0.272 (3.040)0.049 (0.610)0.107 (1.270)0.102 (1.220)
Tertiary education−0.096 (0.620)0.001 (0.010)−0.028 (0.290)−0.145 (1.700)−0.176 (2.140)0.005 (0.060)−0.008 (0.080)
R20.0060.3170.3880.6320.6470.6470.649
Female patients
Secondary education−0.047 (0.310)−0.174 (1.810)−0.084 (0.850)−0.089 (1.050)−0.034 (0.460)0.044 (0.590)0.066 (0.900)
Tertiary education0.209 (1.470)0.209 (2.250)0.208 (2.180)−0.082 (0.970)−0.283 (3.680)−0.030 (0.370)−0.026 (0.330)
R20.0050.2830.3600.5920.6070.6070.609
Fixed effects
Birth yearxxxxxxx
Main diagnosis x     
Main diagnosis × secondary diagnoses  x    
Main diagnosis × secondary diagnoses × procedures   xxxx
Local hospital    xxx
Travel distance × age cuts     xx
Choice variables
Hospital not local hospital      x
University hospital fixed effects      x

For both men and women, we find that the results are sensitive to the controls for medical conditions. Without controls, regression (1), or with controls for diagnoses only, regressions (2) and (3), we find small and insignificant differences between men with primary education and men with tertiary education, whereas men with secondary education wait significantly longer than the other two groups. With maximum controls for medical conditions, regression (4), we find that men with tertiary education wait 14.5% shorter than men with primary education and 33% shorter than men with secondary education.18

For male patients, the relation between education level and waiting time across local hospitals is nonlinear. Local hospitals with a high share of male patients with either tertiary education or primary education have shorter waiting times than local hospitals with a high share of male patients with secondary education. Therefore, when local hospital fixed effects are included, regression (5), the difference between men with tertiary education and men with secondary education decreases to 21%. The difference between men with tertiary education and men with primary education increases to 17.6%.

The effect of education basically disappears when we control for traveling distance. The reason is that the waiting time of male patients increases, whereas the education level decreases in travel distance to the closest hospital.19

As can be seen from Table 3, there are small differences in education level between male patients treated at local hospitals, other hospitals, and university hospitals. This explains why patient choices and treatments at university hospitals have only a modest effect on the male education gradient in waiting time, regression (7). Hence, differences in waiting time between education groups are mainly due to local hospital and travel distance.

The waiting time of women with tertiary education is longer than for the other educational groups without maximum controls for medical condition, regressions (1)–(3). With maximum controls, regression (4), differences in waiting time between the three groups are small and insignificant. When local hospital fixed effects are included (regression (5)), we find an educational gradient in favor of women with tertiary education: women with tertiary education wait 28% shorter than women with primary education and 25% shorter than women with secondary education. The difference between the results in columns (4) and (5) reflects that local hospitals with a high share of female patients with tertiary education have on average longer waiting time than other local hospitals.20 Educational differences in waiting time almost disappear when we control for travel distance (regression (6)). As with male patients, the education level of female patients decreases in distance from hospital.21 Because patients with long travel distance wait longer than other patients, controlling for travel distance makes the education gradient in waiting time flatter. Patient choices and treatments at university hospital have a small effect on the education gradient, the reason being that there are no large educational differences in the type of hospital where treatment takes place (Table 3).

Sensitivity analyses

To check the robustness of the conclusions, we have looked into four possible sources that might bias the estimates. The sensitivity analyses are shown for regressions (4) and (5) for income (Table 5) and regressions (4), (5), and (7) for education (Table 6). The first column of Tables 5 and 6 replicates the results of Tables 2 and 4. Again, we present the results for men and women separately.

Table 5. Effect of income on log (1 + waiting time), sensitivity analysis
Regression abcdef
  1. (a) Replication of results in Table 2; (b) patient episodes referred to hospital treatment 1/1/2004–30/6/2004; (c) without patients from five largest municipalities/parts of city; (d) alternative identification of follow-up treatment; (e) separate diagnoses/procedures fixed effects for inpatient treatment, day treatment, and outpatient treatment; (f) without the 10% of cells with highest standard deviation of income.

Male patients
44th income quintile0.0630 (2.6600)0.0640 (2.3400)0.0510 (1.7200)0.0720 (2.9100)0.0620 (2.6700)0.0520 (1.7300)
3rd income quintile0.9900 (4.3000)0.1190 (4.5600)0.1160 (4.1100)0.1020 (4.2800)0.1000 (4.3900)0.1080 (4.2900)
2nd income quintile0.1560 (6.7000)0.1690 (6.4400)0.1710 (6.4000)0.1540 (6.4200)0.1590 (6.8700)0.1740 (6.8400)
Lowest income quintile0.2350 (11.0300)0.2550 (10.5300)0.2540 (10.2700)0.2350 (10.6600)0.2370 (11.2400)0.2540 (10.8400)
54th income quintile−0.0230 (1.2700)−0.0310 (1.4400)−0.0280 (1.3300)−0.0210 (1.1200)−0.0230 (1.2900)−0.0250 (1.0600)
3rd income quintile−0.0260 (1.3700)−0.0280 (1.2500)−0.0320 (1.4600)−0.0200 (1.0500)−0.0260 (1.3700)−0.0350 (1.6500)
2nd income quintile−0.0400 (1.9900)−0.0470 (2.0200)−0.0600 (2.7900)−0.0360 (1.7900)−0.0390 (1.9500)−0.0440 (1.9400)
Lowest income quintile−0.0090 (0.4200)−0.0210 (0.9000)−0.0230 (1.0500)−0.0030 (0.1400)−0.0080 (0.4100)−0.0120 (0.5200)
Female patients
44th income quintile−0.0760 (2.7400)−0.0689 (2.3600)−0.1020 (3.1700)−0.0920 (3.2500)−0.0760 (2.7200)−0.0720 (1.7000)
3rd income quintile−0.0060 (0.2300)0.0180 (0.6700)−0.0360 (1.2700)−0.0240 (0.9000)−0.0030 (0.1100)−0.0050 (0.1200)
2nd income quintile0.0610 (2.4300)0.0970 (3.7600)0.0500 (1.8500)0.0450 (1.7600)0.0650 (2.5900)0.0620 (1.5300)
Lowest income quintile0.1100 (4.5100)0.1350 (5.3800)0.0890 (3.4200)0.0920 (3.7100)0.1130 (4.6300)0.1110 (2.7700)
54th income quintile−0.0390 (1.6800)−0.0450 (1.6900)−0.0330 (1.4100)−0.0360 (1.6200)−0.0380 (1.6200)−0.0740 (3.4100)
3rd income quintile−0.0090 (−0.3800)−0.0090 (−0.3400)−0.0310 (1.2700)−0.0060 (0.2800)−0.0070 (0.2800)−0.0400 (1.7200)
2nd income quintile−0.0001 (0.0040)0.0080 (0.3000)−0.0200 (0.8100)0.0030 (0.1400)0.0030 (0.1100)−0.0310 (1.3200)
Lowest income quintile0.0220 (1.3800)0.0280 (1.0300)0.0110 (0.4500)0.0350 (1.5100)0.0350 (1.4600)0.0020 (0.1000)
Table 6. Effect of educational achievement on log (1 + waiting time), sensitivity analysis
Regression abcde
  1. (a) Replication of results reported in Table 4; (b) Patient episodes referred to hospital treatment 1/1/2004–30/6/2004; (c) without patients from five largest municipalities/parts of city; (d) alternative identification of follow-up treatment; (e) separate diagnoses/procedures fixed effects for inpatient treatment, day treatment, and outpatient treatment.

Male patients
4Secondary education0.272 (3.040)0.348 (3.380)0.242 (2.560)0.256 (2.810)0.264 (2.960)
Tertiary education−0.145 (1.700)−0.066 (0.680)−0.078 (0.840)−0.118 (1.360)−0.166 (1.950)
5Secondary education0.049 (0.610)−0.082 (0.870)−0.008 (0.100)0.056 (0.700)0.039 (0.490)
Tertiary education−0.176 (2.140)−0.099 (1.030)−0.096 (1.070)−0.171 (2.040)−0.190 (2.310)
7Secondary education0.102 (1.220)0.142 (1.450)0.059 (0.680)0.099 (1.170)0.093 (1.120)
Tertiary education−0.008 (0.080)0.069 (0.660)0.055 (0.570)−0.029 (0.320)−0.022 (0.240)
Female patients
4Secondary education−0.089 (1.050)−0.004 (0.040)−0.157 (1.770)−0.125 (1.460)−0.090 (1.070)
Tertiary education−0.082 (0.970)−0.026 (0.280)0.009 (0.100)−0.072 (0.850)−0.091 (1.090)
5Secondary education−0.034 (0.460)−0.026 (0.320)−0.063 (0.840)−0.052 (0.730)−0.041 (0.570)
Tertiary education−0.283 (3.680)−0.305 (3.560)−0.217 (2.700)−0.302 (3.950)−0.291 (3.800)
7Secondary education0.066 (0.900)0.068 (0.840)0.028 (0.380)0.041 (0.560)0.058 (0.810)
Tertiary education−0.026 (0.330)−0.058 (0.640)0.010 (0.120)−0.053 (0.670)−0.034 (0.420)

The first sensitivity check we perform is whether treatments with exceptionally long waiting time affect the results. In the sensitivity analysis, we change the date of referral to include all treatments with up to 18 months waiting time. The results are presented in column (b) in Tables 5 and 6. We notice that the estimated effects of income in regression (4) become slightly stronger when the referral period is shortened. Because the number of excluded patient episodes is higher the longer the time of referral, this may indicate that the results presented in Table 2 underestimate the effect of income. Quantitatively, there are however no reasons to believe that the bias is large. Concerning education, we observe different effects of changing the referral period for men and women in regression (5). For men, the effect of tertiary education becomes weaker, whereas the effect of tertiary education becomes slightly stronger for women.

Because we cannot follow patients over time and patients that move between hospitals, we have assumed that two patient episodes with the same date of referral, sex, age, place of residence (municipality or part of city), and main diagnosis relate to the same patient. The consequences of this assumption are that some patient episodes that are included might be following up consultations for patients that already are treated and that some patient episodes are excluded from the data set even if they are the initial treatment. In the sensitivity analyses, we remove patients from the five largest municipalities/parts of city that have the highest number of treatments (because the problem will be more severe in large municipalities).22 We also add the criterion that two patient episodes also must include the same secondary diagnoses (in addition to the five criteria mentioned earlier, before a patient episode is dropped). These results are presented in columns (c) and (d) in Tables 5 and 6. We conclude that the results are not very sensitive to the way patient episodes are defined.

We also check whether the results are sensitive to the way patients are admitted to hospitals. The issue is that differences in the medical conditions of patients may be correlated with the way patients are admitted, even though we control thoroughly for diagnoses and procedures. Hence, we include separate fixed effects for medical conditions for the three different ways of admitting patients: inpatient treatment, day treatment, and outpatient treatment. The results are presented in column (e) in Tables 5 and 6. In these regressions, we thus control for all possible combinations of the way patients are admitted, main diagnosis, secondary diagnoses, and procedures. We notice that the results are not sensitive to the way patients are admitted to hospitals.

To check whether within-cell dispersion in income may bias our results, we have omitted patients in population cells with high income dispersion, measured by either the within-cell standard deviation of income or the within-cell difference between the 90th and 10th income percentiles. We find that our results are not much affected. Column (f) in Table 5 shows results without cells with the 10% highest standard deviation of income; alternative cutoff values as well as reduction of the sample based on the 90th–10th percentile income difference give similar results.

Finally, and as an alternative to the OLS specification, we have estimated some simple duration models of the relationship between socioeconomic status and waiting time. Including a full set of fixed effects for medical condition (either of the three fixed effects alternatives) in a duration analysis would require computer resources that vastly exceed ours. However, comparison of OLS and duration analyses with parsimonious specifications indicate that the results are very similar. This is to be expected because the combination of LOG waiting times and OLS is approximately equivalent to a basic duration model (proportional exponential hazard rate).23,24

CONCLUDING REMARKS

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. INSTITUTIONAL FEATURES
  5. DATA
  6. METHODOLOGY AND EMPIRICAL ANALYSIS
  7. CONCLUDING REMARKS
  8. REFERENCES

In this paper, we have investigated whether socioeconomic status, measured by income and education, affects waiting time when we control for patients' medical condition (severity of illness) and hospital-specific effects such as local hospital and distance to hospital. When we investigate the sample of all elective patients, we find evidence of discrimination with regards to both income and education when only controls for severity of illness are included. These effects are quite large, as men with tertiary education wait about 14.5% shorter than men with only primary education. For women, evidence of discrimination is found with regards to income as women in the lowest income quintile wait 11% longer than women within the highest income quintile. However, when controls for hospital-specific factors such as attachment to local hospital, traveling time, or choice of hospital are included, we find very little evidence of discrimination. Specifically, we find that local hospital explains most of the income gradient. The reason is that hospitals located in low-income regions have longer waiting time than hospitals located in high-income and middle-income regions. Hence, controlling for local hospitals makes the income gradient flatter. Inclusion of travel distance further weakens the association between income and waiting time. This follows as patients' income decreases in traveling distance whereas waiting time increases in this metric. Finally, we check how treatments at university hospitals impact on the income gradient. What we find is that the impact is limited. This is a consequence of very little systematic income difference between patients treated in different hospital types.

When we analyze the impact of hospital-specific factors on the educational differences in waiting time, we find that the differences are mainly due to variation between educational groups in local hospital and travel distance for men and travel distance for women. For male patients, the relation between education level and waiting time across local hospitals is nonlinear. Local hospitals with a high share of male patients with either tertiary education or primary education have shorter waiting times than local hospitals with a high share of male patients with secondary education. Therefore, the differences between men with tertiary education and secondary education decreases, whereas the difference between men with tertiary education and primary education increases when hospital fixed effects are included. However, the effect of education basically disappears when we control for traveling distance. The reason is that the waiting time of male patients increases, whereas the education level decreases in travel distance to the closest hospital. For women, we find an educational gradient in favor of women with tertiary education when local hospital fixed effects are included. But again, this effect disappears when travel distance is included. The effect is similar to the one we observe for men: the educational level decreases the longer travel distance, and those who live closer to the hospital experience shorter waiting time. Hence, the educational gradient becomes flatter when travel distance is controlled for. Our results are insensitive to the robustness checks we perform.

One of the novelties of our analysis is that the data allow us to include very extensive controls for patients' severity. This is important because, according to the prioritization guidelines often used in many health systems of the National Health Service type, more severely ill patients should have shorter waiting times (Siciliani and Hurst, 2005; Gravelle and Siciliani, 2008). From the results, we see that controlling for severity is important for both women and men, but it seems that the exact way of controlling for severity matters more for women.

The fact that a socioeconomic gradient might give unacceptable equitable consequences raises the question of why individuals experience longer waiting time. What we have performed is to examine whether hospital factors explain the socioeconomic differences in waiting time or if the socioeconomic gradient still remains when factors such as attachment to local hospital and distance to hospital are controlled for. That is, we do not try to explain inequalities with patient-level variables. Such an approach would require not only that we include income and education in the same regression but also that we have access to data on other variables that describe a patient's personal resources and that are correlated with income and/or education. Candidates for such variables are attachment to the labor force (employed, unemployed, or on social security) and family relations, just to mention two. In addition, one has to tackle the problem of reversed causality; the more treatment one receives, the healthier one is and the easier it is to obtain income and, for students, education. Instead, our approach is to focus on two variables that are commonly used to describe inequality, check how they are related to waiting time, and try to explain any differences by hospital-specific conditions relative to where patients' live.

One important contribution of our work is that we investigate whether inequalities arise across hospitals or within the hospital. With respect to income, we find evidence of inequalities across hospitals, but not within. With respect of education, we find evidence of inequality within hospitals, but not across. The latter results are also found in a study based on England (Laudicella et al., 2012). Both these studies illustrate the importance of controlling for hospital fixed effects when analyzing inequalities, as sometimes omitting hospital effects underestimates the socioeconomic inequality in waiting times.

A limitation with the approach we have used in this paper is that we cannot match socioeconomic status at individual level with administrative patient data.25 Instead, we are linking patient-level data from administrative databases with data on income and education from population cells that combine gender, age, and municipality. In this respect, we follow a recent trend that makes use of administrative data to investigate inequalities in health or health care. We have chosen to focus on elective care, and acute care is therefore not included in our analyses. This might potentially bias out results, especially if it is the case that patients with lower socioeconomic status are more likely to receive acute treatment. One reason for this might be that these patients have longer waiting time and thus may become acute ill while waiting. But because we do not find much evidence of discrimination, we do not think our results are sensitive to the exclusion of acute care treatment. Finally, one might argue that household income is a better proxy for socioeconomic status than individual income. This might especially be the case for married women that are not active in the labor market. Our data are however not rich enough to shed light on this matter.

  1. 1

    We do not intend to explain whether any socioeconomic inequality in waiting time is caused by a person's socioeconomic status or any personal variables that are correlated with socioeconomic status. Such an endeavor requires a much richer data set than we have access to.

  2. 2

    The Carstairs index of deprivation is a composite deprivation index based on car ownership, unemployment, overcrowding, and social class within output areas. It is calculated by the Office of National Statistics (e.g., Morgan and Baker, 2006).

  3. 3

    There is a broad literature measuring equality in healthcare utilization, van Doorslaer and Wagstaff (2000). This literature tests whether individuals with higher socioeconomic status have higher utilization (as measured by number of visits), after controlling for need (self-reported health). The evidence broadly suggests prorich inequality for physician visits. When visits are split between specialist visits and family doctor consultations in gatekeeping systems, the evidence suggests prorich inequity for the former and propoor inequity for the latter (van Doorslaer et al., 2004). Grasdal and Monstad (2009) analyzed and compared inequality in use of physician visits in Norway on the basis of survey data. For specialist services, they found prorich inequality in the probability of seeing an outpatient specialist. There is a limited literature that makes use of administrative data to investigate inequalities in health and health care (on length of stay: Cookson and Laudicella, 2009; on health care utilization: Propper et al., 2005; on prioritization and patients' rights: Carlsen and Kaarboe, 2010).

  4. 4

    Hagen and Kaarboe (2006) and Magnussen et al. (2007) provided more detailed descriptions of the Norwegian hospital sector.

  5. 5

    We will refer to category (i) hospitals as ‘local hospitals’, category (ii) hospitals as ‘other hospitals’, and category (iii) hospitals as ‘university hospitals’.

  6. 6

    The statutory retirement age in Norway is 67 years.

  7. 7

    Income is annual pretax income from employment, self-employment, and transfers (pensions, social assistance benefits, etc.). Capital income is not included because the administrative registers of Statistics Norway lack data about capital gains.

  8. 8

    The data period is chosen to match the data on income and education. Some patients will wait longer than 15 months and receive treatment in 2006. But because only 4% of the patients referred in this period wait longer than 1 year, we believe that only a few patient stays are excluded by design.

  9. 9

    Our rational for focusing on patients with a referral date between 1 January 2004 and 31 August 2004 is a change in the law of Patients' Rights introduced on 1 September 2004 (Ministry of Health and Social Services, 1999, 2003). The law may have changed the prioritization practice of the hospitals. Askildsen et al. (2010, 2011) and Januleviciute et al. (2010) analyzed the effects of the reform. The authors found very little effect of the reform.

  10. 10

    In Section 4.1, we check whether the results are sensitive to this assumption.

  11. 11

    There is a positive but modest correlation between cell size and income dispersion.

  12. 12

    The ICD provides codes to classify diseases and a wide variety of signs, symptoms, abnormal findings, complaints, and external causes of injury or diseases. Norway uses the ICD-10 version to classify all hospital stays.

  13. 13

    Some of the groups only include one patient, so the effective number of observations that are used to identify the relationship between waiting time and socioeconomic status is lower, but still above 200,000 (270,000) for men (women).

  14. 14

    Sixty-two hospitals are local hospital for at least one combination of diagnosis, sex, and municipality.

  15. 15

    Number of patients is used as weights.

  16. 16

    The adjusted average waiting time for male (female) patients that live more than 50 km away from the closest hospital is 93 days (98 days) versus 77 days (82 days) for patients that live less than 50 km away. The share of male (female) patients in the top cell income quintile is 0.15 (0.18) if distance is less than 50 km and 0.05 (0.04) otherwise.

  17. 17

    For male patients, the income distribution of the two types of patients is similar. For female patients, other hospitals have a larger share of patients in the highest and the two lowest cell income quintiles, whereas local hospitals have a higher share of patients in the second and third quintiles (Table 3).

  18. 18

    0.417 (=0.272 − (−0.145)) is 33% of 1.272.

  19. 19

    The average cell population share with tertiary education is 28% for male patients living closer than 50 km from a hospital and 16% for male patients living more than 50 km from a hospital.

  20. 20

    The weighted correlations across local hospitals between average adjusted waiting time and the share of female patients in the three education groups are 0.02 for primary education, −0.12 for secondary education, and 0.08 for tertiary education.

  21. 21

    The average cell population share with tertiary education is 30% for female patients living closer than 50 km from a hospital and 22% for female patients living more than 50 km from a hospital.

  22. 22

    The average number of inhabitants per municipality in 2006 is 8296 when the five largest municipalities/parts of city are excluded.

  23. 23

    We thank Luigi Siciliani for pointing out this result.

  24. 24

    The results are available upon request.

  25. 25

    We are not aware of any study that has access to such data.

REFERENCES

  1. Top of page
  2. ABSTRACT
  3. INTRODUCTION
  4. INSTITUTIONAL FEATURES
  5. DATA
  6. METHODOLOGY AND EMPIRICAL ANALYSIS
  7. CONCLUDING REMARKS
  8. REFERENCES
  • Askildsen JE, Holmås TH, Kaarboe O. 2010. Prioritization and patients' rights: analysing the effect of a reform in the Norwegian Hospital Sector. Social Science & Medicine 70(2): 199208.
  • Askildsen JE, Holmås TH, Kaarboe O. 2011. Monitoring prioritisation in the public health care sector by use of medical guidelines. The case of Norway. Health Economics 20(8): 958970.
  • Carlsen F, Kaarboe O. 2010. Norwegian priority guidelines: estimating the distributional implications across age, gender and SES. Health Policy 95: 26470.
  • Cookson R, Laudicella M. 2009. Do the poor still cost more? The relationship between small area income deprivation and length of stay for elective hip replacement in the English NHS from 2001/2 to 2006/7. Health Economics and Data Group (HEDG), University of York Working Paper 09/07.
  • Cooper ZN, McGuire A, Jones S, Le Grand J. 2009. Equity, waiting times, and NHS reforms: retrospective study. British Medical Journal 339: b3264.
  • Grasdal A, Monstad K. 2009. Inequity in the use of physician services in Norway. Changing patterns over time? Department of Economics, University of Bergen Working Paper 05/09.
  • Gravelle H, Siciliani L. 2008. Is waiting-time prioritisation welfare improving? Health Economics 17(2): 167184.
  • Hagen TP, Kaarboe O. 2006. The Norwegian hospital reform of 2002: central government takes over ownership of public hospitals. Health Policy 76: 32033.
  • Januleviciute J, Askildsen JE, Holmås TH, Kaarboe O, Sutton M. 2010. The impact of different prioritisation policies on waiting times: different prioritisation policies. A comparative analysis for Norway and Scotland. Working paper 07/10, Department of Economics, University of Bergen.
  • Laudicella M, Siciliani L, Cookson R. 2012. Waiting times and socio-economic status: evidence from England. Social Science & Medicine 74: 13311341.
  • Magnussen J, Hagen TP, Kaarboe O. 2007. Centralized or decentralized? A case study of Norwegian hospital reform. Social Science & Medicine 64: 21292137.
  • Ministry of Health and Social Services. 1999. Lov om pasientrettigheter. Innstilling O. nr. 91 1998–99. (Patient Rights Act).
  • Ministry of Health and Social Services. 2003. Endring i lov 2. juli 1999 nr. 63 om pasientrettigheter. Besl. O. nr 23 2003–2004. (Amendment to Patient Rights Act).
  • Monstad K. 2007. Patients' preferences for choice of hospital. Chapter 1 in Doctoral Thesis, The Norwegian School of Economics and Business Administration. http://bora.nhh.no
  • Morgan O, Baker A. 2006. Measuring deprivation in England and Wales using 2001 Carstairs scores. Health Statistics Quarterly 31: 2833.
  • Propper C, Eachus J, Chan P, Pearson N, Smith GD. 2005. Access to health care resources in the UK: the case of care for arthritis. Health Economics 14: 391406.
  • Siciliani L, Hurst J. 2005. Tackling excessive waiting times for elective surgery: a comparison of policies in 12 OECD countries. Health Policy 72(2): 201215.
  • Siciliani L, Verzulli R. 2009. Waiting times and socio-economic status among elderly Europeans: evidence from SHARE. Health Economics 18: 12951306.
  • Van Doorslaer E, Wagstaff A. 2000. Equity in health care financing and delivery. In Handbook of Health Economics, Culyer AJ, Newhouse JP (eds.), Elsevier Science, North-Holland: Amsterdam.
  • Van Doorslaer E, Koolman X, Jones AM. 2004. Explaining income-related inequalities in doctor utilization in Europe. Health Economics 13: 629647.
  • Vrangbæk K, Ostergren K, Birk HO, Winblad U. 2007. Patients' reactions to hospital choice in Norway, Denmark and Sweden. Health Economics, Policy, and Law 2: 125152.