JECH Online First, published on October 19, 2010 as 10.1136/jech.2009.

104554 Research report

Contribution of income-related inequality and healthcare utilisation to survival in cancers of the lung, liver, stomach and colon
Jun Yim,1 Seung-sik Hwang,2 Keun-young Yoo,3 Chang-yup Kim4
1 Department of Preventive Medicine, Gachon University of Medicine and Science, Incheon, Korea 2 Department of Social and Preventive Medicine, Inha University School of Medicine, Incheon, Korea 3 Department of Preventive Medicine, Seoul National University College of Medicine, Seoul, Korea 4 School of Public Health, Seoul National University, Seoul, Korea

Correspondence to Jun Yim, Department of Preventive Medicine, Gachon University of Medicine and Science, 534-2 Yeonsu3-dong, Yeonsu-gu, Incheon, 406-799, Korea; yim99@gachon.ac.kr Accepted 12 June 2010

ABSTRACT Objectives To examine differences in the survival rates of cancer patients according to socioeconomic status, focusing on the role of the degree of healthcare utilisation by the patient. Methods An observational follow-up study was done for 261 lung cancer, 259 liver cancer, 268 stomach cancer and 270 colon cancer patients, diagnosed during 1999e2002. Income status and healthcare utilisation were assessed with National Health Insurance (NHI) data; survival during 1999e2002 was identied by death certicate. HRs and 95% CI were derived from Cox proportional hazards regression. Results and Conclusions The HRs for low income status are larger for colon cancer (2.37, 95% CI 1.17 to 4.80), followed by stomach (1.67, 95% CI 1.01 to 2.78), liver (1.57, 95% CI 1.03 to 2.39) and lung cancers (1.46, 95% CI 0.99 to 2.14). In the model including the variable of healthcare utilisation, colon and stomach cancers exhibited a lower HR in the moderate healthcare utilisation groups (0.40, 95% CI 0.21 to 0.76 in colon; 0.59, 95% CI 0.37 to 0.96 in stomach), whereas for liver cancer, the high utilisation group exhibited a higher hazard (1.72, 95% CI 1.07 to 2.75). A lower income status is independently related to a shorter survival time in cancer patients, especially in less fatal cancers. Healthcare utilisation independently affects the likelihood of survival from colon and stomach cancers, implying that a moderate degree of healthcare utilisation contributes to a longer survival time.

survival among different SES groups. Various factors have been proposed as possible underlying reasons for this SES effect: stage at diagnosis, treatment modality, quality of treatment, host factors and psychosocial factors.8 Among them, it has been accepted that stage of cancer at diagnosis is probably the most important contributing factor to the different rates of survival among the different SES groups, but the contribution of quality and mode of treatment and psychosocial factors has scarcely been studied. Moreover, very little attention has been paid to the degree of healthcare utilisation compared to other treatment factors such as modality and quality. Therefore, the aim of this study was to analyse differences in the survival rates of cancer patients according to SES, focusing on the role of the degree of healthcare utilisation by the patient. We hypothesised that patients with a lower SES would utilise healthcare less and have a shorter survival time.

METHODS Data
All cases registered during 2000 were initially sampled from the ve major hospitals that had the largest number of registered cancer patients nationally in the Korean National Cancer Registry. For each cancer, a cohort of 300 patients was sampled from the ve hospitals using stratied random sampling according to the total number of patients, and a survey of their medical records was conducted by four trained medical record administrators. Patients diagnosed prior to 1999 were excluded and the nal analysis was carried out on 261 cases with lung cancer, 259 with liver cancer, 268 with stomach cancer and 270 with colon cancer. Morphology of lung cancer was composed of squamous cell carcinoma, not otherwise specied (NOS) (32.8%), adenocarcinoma, NOS (32.1%) and small cell carcinoma, NOS (11.5%). Morphology of liver cancer was composed of hepatocarcinoma (79.5%), cholangiocarcinoma (8.1%) and neoplasm, NOS (6.6%). Morphology of stomach cancer was composed of adenocarcinoma, NOS (50.4%), signet ring cell carcinoma (16.0%) and tubular adenocarcinoma (13.8%). Morphology of colon cancer was composed of adenocarcinoma, NOS (75.4%), tubular adenocarcinoma (7.0%) and neoplasm, NOS (5.5%). The patients general information and risk factors, including Karnofsky performance status scale (KPSS) for physical performance,9 were
1 of Publishing Group Ltd under licence. 4

INTRODUCTION
Cancer incidence and mortality have increased sharply in South Korea (Korea) in recent years. The fraction of deaths attributable to cancer has more than doubled from 10.5% of the total number of deaths in 1981 to 24% in 2000.1 As in many developed countries, cancer has emerged as one of the most critical health problems in Korea. However, the burden of cancer is not equal among all population groups. It is already known that socioeconomic status (SES) affects the mortality rate from cancersdmortality is particularly high in patients with low incomes.2e7 The overall mortality rate due to a fatal disease is determined by the occurrence of the disease and the survival rate. However, the effect of SES on the survival rate of cancer patients has been studied less thoroughly than has the occurrence of the disease,8 although such effects have been noted for most cancer types and for several countries. In particular, there are only a few studies that have dealt specically with the aetiology of different cancer

Yim J, Hwang S-s, Article author (or Community Health (2010). 2010. Produced by BMJ Copyright Yoo K-y, et al. J Epidemiol their employer) doi:10.1136/jech.2009.104554

Research report
extracted from their medical records, and the cancers were staged according to the summary staging of the surveillance, epidemiology and end results programme (SEER) of the US National Cancer Institute.10 The death or survival of patients was conrmed by the 1999e2002 death certicate statistics of the Korean National Statistical Ofce. The maximum observation period was 48 months. Table 1
Variable Number of patients Type of cancer Lung Liver Stomach Colon Age*, mean (SD) <60 $60 Gender Male Female Stage of cancery Early Advanced Family history Negative (%) Positive (%) Physical performancez Good Poor Outpatient visits per month <1 $1, <3 $3 Hospital A B C D E

Individual baseline characteristics of 1058 cancer patients

Income status High (N, %) 358 (33.8) 91 80 86 101 59.3 170 188 (34.9) (30.9) (32.1) (37.4) (11.9) (31.7) (36.1) Middle (N, %) 436 (41.2) 99 113 114 110 56.5 239 197 (37.9) (43.6) (42.5) (40.7) (11.5) (44.5) (37.8) Low (N, %) 264 (25.0) 71 66 68 59 58.2 128 136 (27.2) (25.5) (25.4) (21.9) (11.4) (23.8) (26.1)

Estimates of income and degree of healthcare utilisation

Since the payroll of the employed and the means-tested income of the self-employed determines the National Health Insurance (NHI) premium paid by individuals (the level of premium thus reecting the income status of the insured), the income status of patients could be obtained from their NHI insurance prole. The estimated income level was quartered from the total population registered with NHI and Medicaid, but for the analysis, income level was divided into three groups, high, middle and low, after the interquartile range of estimated income levels had been integrated into the middle group. The population of Medicaid was included in the low-income group. Every healthcare institution has to submit all claims data to the insurer to be reimbursed, and payment for cancer treatment by private insurance is only supplementary to the NHI, and so almost all healthcare utilisation of cancer patients is included in the NHI database. Information about healthcare utilisation for each patient was therefore obtained from NHI claims data. The degree of healthcare utilisation was dened by the number of outpatient visits per month. Since the number of outpatient visits was signicantly correlated with whether the patient had been admitted or not using the ManteleHaenszel c2 test (p<0.001), outpatient visits per month was selected as a representative variable to indicate the degree of healthcare utilisation.

246 (33.1) 112 (35.7) 96 (37.1) 262 (32.8) 333 (34.0) 25 (31.7) 255 (33.6) 97 (35.1) 76 (30.5) 188 (34.5) 94 (35.6) 36 102 58 127 35 (23.5) (39.7) (37.4) (36.7) (23.8)

312 (41.9) 124 (39.5) 103 (39.8) 333 (41.7) 401 (41.0) 35 (44.3) 310 (40.9) 118 (42.8) 109 (43.8) 232 (42.6) 95 (36.0) 61 97 66 149 63 (39.9) (37.7) (42.6) (43.1) (42.9)

186 (25.0) 78 (24.8) 60 (23.2) 204 (25.5) 245 (25.0) 19 (24.1) 193 (25.5) 61 (22.1) 64 (25.7) 125 (22.9) 75 (28.4) 56 58 31 70 49 (36.6) (22.6) (20.0) (20.2) (33.3)

Statistical analysis
We used the KaplaneMeier method for the unadjusted analyses to examine the differences in the rate of survival from each type of cancer. The survival rate from different income groups was compared using the log-rank test. In addition, for multivariate analysis on the factors affecting survival, Cox proportional hazards regression was performed to calculate the HRs for two models. Variables considered for inclusion were income status, age, gender, cancer stage, KPSS, family history, outpatient visits per month and kind of hospital. The latter was found to be nonsignicant in univariate analysis, and was not included in subsequent models. In model 1, income status and related variables, which included age, gender, cancer stage, KPSS and family history were studied. In model 2, as the full model, outpatient visits were added to model 1. The signicance of adding outpatient visits to model 2 was estimated by the likelihood ratio test, and models for liver and colon cancers showed statistically signicant results (p<0.05).

*p<0.05 for ANOVA. ySummary staging of the surveillance, epidemiology and end results programme (0e1 early, 2e7 advanced). zKarnofsky performance status scale (>70 good, #70 poor).

Table 2 lists the results from the Cox models for the mortality of cancers. The HRs of the low-income group are consistently larger than those of high-income group, in both models 1 and 2. The HRs are larger for colon cancer, followed by stomach, liver and lung cancers, in that order. In model 2, stomach and colon cancers exhibited a signicantly lower hazard in the moderate healthcare utilisation groups ($1, <3 visits), and for liver cancer the high utilisation group ($3 visits) exhibited a signicantly higher hazard.

DISCUSSION Findings
Most of the results of this study are consistent with those of previous studies in which it was shown that a lower socioeconomic status is related to a shorter survival time in cancer patients. The impact of socioeconomic status on the survival of cancer patients is independent of the stage at diagnosis, physical performance, family history and the degree of healthcare utilisation. Consequently, the differences in survival among the different levels of income status cannot be explained by variations in the distribution of these factors among the income status groups. Interestingly, in our study the results show that the effect of having a lower income on the mortality due to cancer is more marked in less fatal cancers such as colon cancer. On the contrary, there is no clear relationship between income status

RESULTS
Table 1 lists the general baseline characteristics of the patients. Except for age, the characteristics of the patients were not signicantly different among the different income statuses. In particular, the degree of healthcare utilisation was not signicantly different among the income groups studied. By the end of 2002, the survival rate was highest for colon cancer, followed by stomach, liver and lung cancers, in that order. The low-income group exhibited a less favourable survival pattern for liver, stomach and colon cancers, but no such difference was observed for lung cancer (gure 1).
2 of 4

Yim J, Hwang S-s, Yoo K-y, et al. J Epidemiol Community Health (2010). doi:10.1136/jech.2009.104554

Research report
Figure 1 KaplaneMeier curves for lung cancer (A), liver cancer (B), stomach cancer (C), and colon cancer (D) among three different income groups: highest, 2nd to 3rd, lowest income group in Korea, 1999e2002. The log-rank test was used for statistical comparisons.

and survival for the more fatal cancers, such as lung cancer. These ndings suggest that survival of less fatal cancers is inuenced more strongly by income-related factors, such as nutrition, housing, level of healthcare utilisation and social support.11 12 Another factor that may explain the different survival times among those with differing SES is the quality of treatment received.8 In this study, we have not explored directly the quality of treatment received by each patient, but each hospital may have a different level of quality and hospital factors could affect the quality of treatment given. Hospital factors, however, were not found to be signicant for unadjusted HRs and were excluded in models 1 and 2. Inconsistent with our hypothesis, the degree of healthcare utilisation was not signicantly different among the different income status groups. This is probably attributable to the universal coverage of the NHI for the entire population, in which basic healthcare utilisation is ensured even for those in the lower-income groups, although there is still a high rate of copayment that sometimes hinders access to healthcare. Moreover, healthcare utilisation independently affects the survival of stomach and colon cancers, implying that a moderate degree of healthcare utilisation contributes to a longer survival period, even after adjusting for income status. Appropriate healthcare utilisation should be ensured in cancer patients to improve survival across all levels of income status. In spite of a partially benecial effect, healthcare utilisation does not exhibit a doseeresponse relationship, the highest level of utilisation being related to a higher hazard than in moderateutilisation groups for all of the cancers studied. These ndings can be explained partly by the fact that some of the healthcare

utilisation was probably the result of an aggravated clinical course for patients rather than being initiated by the patients themselves or their providers in order to contribute to a better outcome. Consequently, if we suggest that appropriate healthcare utilisation is benecial to the survival of cancer patients, cases for which utilisation is a result of a more serious health condition should be excluded from further analyses.

Methodological issues
Our study has some limitations. First, the NHI premium should be justied as an appropriate measure of income status. Since the premium of the employed is based on the payroll and that of the self-employed is based on their estimated income from various sources, the extent to which the actual income level is reected in the premium would be different between these two groups. However, the NHI premium is continuously revised to accurately reect income, and so no other measure was thought to be superior as a proxy for real income. In particular, the result analysed on the population of employed was not different from that of employed and self-employed. Another limitation is that the death certicate statistics of the National Statistical Ofce were not fully validated due to the limited information available on the cause of death. Although most of the data contain some information about the cause of death, it was difcult to establish whether or not the cause was directly related to cancer. Therefore, cancer patients who died of other causes might have been included in the cancer mortality statistics. However, because death from cancer is far more dominant than from other causes,8 this type of inaccuracy is unlikely to alter the results of this study to any large extent. A third limitation is that
3 of 4

Yim J, Hwang S-s, Yoo K-y, et al. J Epidemiol Community Health (2010). doi:10.1136/jech.2009.104554

Research report
Table 2 Adjusted HRs from multivariable Cox models for different mortality according to income status among 1058 cancer patients, 1999e2002
Unadjusted HR (95% CI) Lung Income status High 1.00 Middle 1.17 Low 1.25 Outpatient visit <1 1.00 $1, <3 0.66 $3 0.88 Liver Income status High 1.00 Middle 1.23 Low 1.68 Outpatient visit <1 1.00 $1, <3 0.79 $3 1.41 Stomach Income status High 1.00 Middle 1.68 Low 1.81 Outpatient visit <1 1.00 $1, <3 0.97 $3 1.28 Colon Income status High 1.00 Middle 1.73 Low 1.87 Outpatient visit <1 1.00 $1, <3 0.45 $3 0.92 Model 1* Adjusted HR (95% CI) Model 2y Adjusted HR (95% CI)

What is already known on this subject

< There is a strong association between cancer survival and

socioeconomic status.

ref. (0.84 to 1.65) (0.87 to 1.80) ref. (0.44 to 0.97) (0.59 to 1.31)

1.00 ref. 1.07 (0.76 to 1.52) 1.48 (1.01 to 2.17)

1.00 ref. 1.11 (0.78 to 1.57) 1.46 (0.99 to 2.14) 1.00 ref. 0.73 (0.48 to 1.11) 0.94 (0.62 to 1.41)

What this study adds

< The effect of having a lower income on the mortality due to

ref. (0.84 to 1.80) (1.11 to 2.53) ref. (0.55 to 1.13) (0.90 to 2.21)

1.00 ref. 1.23 (0.84 to 1.80) 1.54 (1.01 to 2.36)

1.00 ref. 1.27 (0.87 to 1.87) 1.57 (1.03 to 2.39) 1.00 ref. 0.93 (0.65 to 1.34) 1.72 (1.07 to 2.75)

cancers is more marked in less fatal cancers such as colon cancer. < Appropriate healthcare utilisation contributes to longer survival times, regardless of income status.

Therefore, we do not believe that this limitation is critical with regard to the formulation of the data pool.

Conclusions
A lower income status is independently related to a shorter survival time in cancer patients, particularly in less fatal cancers. Healthcare utilisation independently affects the survival of colon and stomach cancers, implying that a moderate degree of healthcare utilisation contributes to a longer survival period.
Funding Ministry of Health and Welfare, Republic of Korea. Competing interests None declared. Provenance and peer review Not commissioned; externally peer reviewed.

ref. (1.07 to 2.62) (1.11 to 2.96) ref. (0.61 to 1.55) (0.76 to 2.17)

1.00 ref. 1.58 (1.01 to 2.47) 1.69 (1.02 to 2.79)

1.00 ref. 1.58 (1.01 to 2.49) 1.67 (1.01 to 2.78) 1.00 ref. 0.59 (0.37 to 0.96) 0.87 (0.50 to 1.54)

ref. (0.99 to 3.04) (0.99 to 3.53) ref. (0.25 to 0.81) (0.51 to 1.65)

1.00 ref. 2.30 (1.24 to 4.28) 2.44 (1.20 to 4.97)

1.00 ref. 2.31 (1.25 to 4.29) 2.37 (1.17 to 4.80) 1.00 ref. 0.40 (0.21 to 0.76) 0.85 (0.45 to 1.59)

REFERENCES
1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. 12. KNSO. Annual report on the cause of death statistics. Seoul: Korea National Statistical Ofce 2001. [Korean]. Kaplan GA, Pamuk ER, Lynch JW, et al. Inequality in income and mortality in the United States: analysis of mortality and potential pathways. BMJ 1996;312:999e1003. Kennedy BP, Kawachi I, Prothrow-Stith D. Income distribution and mortality: cross sectional ecological study of the Robin Hood Index in the United States. BMJ 1996;312:1004e7. Lynch JW, Kaplan GA, Pamuk ER, et al. Income inequality and mortality in metropolitan areas of the United States. Am J Public Health 1998;88:1074e80. Blakely T, Atkinson J, Kiro C, et al. Child mortality, socioeconomic position, and one-parent families: independent associations and variation by age and cause of death. Int J Epidemiol 2003;32:410e18. Wong MD, Shapiro MF, Boscardin WJ, et al. Contribution of major diseases to disparities in mortality. N Engl J Med 2002;347:1585e92. Singh GK, Miller BA, Hankey BF, et al. Changing area socioeconomic patterns in U.S. cancer mortality, 1950-1998: Part IeAll cancers among men. J Natl Cancer Inst 2002;94:904e15. Auvinen A, Karjalainen S. Possible explanations for social class differences in cancer patient survival. In: Kogevinas M, Pearce N, Susser M, et al, eds. Social inequalities and cancer. Lyon: International Agency for Research on Cancer 1997:377e97. Karnofsky DA, Burchenal JH. The clinical evaluation of chemotherapeutic agents in cancer. In: Macleod CM, ed. Evaluation of chemotherapeutic agents. New York NY: Columbia University Press, 1949:199e205. Ries LAG, Eisner MP, Kosary CL, et al. SEER cancer statistics review, 1973e1999. Bethesda, MD: National Cancer Institute, 2002. Kravdal O. Children, Family and cancer survival in Norway. Int J Cancer 2003;105:261e6. Kawachi I, Kennedy BP, Lochner K, et al. Social capital, income inequality, and mortality. Am J Pub Health 1997;87:1491e8.

*Model for income status; age; gender (male, female); summary staging of the surveillance, epidemiology, and end results programme (SEER) (0e1 early, 2e7 advanced); Karnofsky performance status scale (KPSS) (>70 good, #70 poor); and family history (no, yes). yModel for variables in model 1 and outpatient visits per month.

area-based measures of SES were not available. These would have allowed us to evaluate independently the area and individual effects using multilevel analysis. Moreover, we could not exclude cases with serious health conditions, which probably induced high healthcare utilisation, due to the limitation of medical records. The nal limitation is that only patients registered in ve hospitals were sampled. Consequently, it is difcult to represent a tendency of the total population. If this problem could be solved, we would have to consider the relative survival of total population with cancer, but have not done so. However, the total number of cancer patients who are registered in these hospitals represents more than 25% of the total number of cancer patients registered in hospitals throughout Korea.

4 of 4

Yim J, Hwang S-s, Yoo K-y, et al. J Epidemiol Community Health (2010). doi:10.1136/jech.2009.104554