Japanese Journal of Clinical Oncology Advance Access published online on July 16, 2007
Japanese Journal of Clinical Oncology, doi:10.1093/jjco/hym047
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
© 2007 Foundation for Promotion of Cancer Research
Cancer Survival Trends in Osaka, Japan: the Influence of Age and Stage at Diagnosis
1 Graduate School of Medicine, Osaka University, Osaka, Japan
2 London School of Hygiene and Tropical Medicine, London, UK
3 Osaka Medical Center for Cancer and Cardiovascular Diseases, Osaka, Japan
For reprints and all correspondence: Yuri Ito, Osaka Medical Center For Cancer and Cardiovascular Disease, 1-3-3 Nakamichi, Higashinari-ku, Osaka, 537-8511, Japan. E-mail: itou-yu2{at}mc.pref.osaka.jp
Received January 3, 2007; accepted February 3, 2007
 |
Abstract |
|---|
 TOP Abstract INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
Background: Five-year survival is an important index for evaluating the overall effectiveness of cancer diagnosis and treatment. The aim of this study was to monitor trends in 5-year cancer survival in Osaka, Japan, during 1975–94, with adjustment for trends in age and stage at diagnosis.
Methods: Five-year crude survival was estimated for cancers of the stomach (39 697 patients), lung (17 400 patients) and breast (11 988 women) in four consecutive 5-year periods, with adjustment for age and/or stage, using the age and stage distribution of cases diagnosed during 1975–79 as standard weights. We estimated the impact of adjusting for either age or stage alone by their proportionate impact on the trend in fully (age- and stage-) adjusted survival.
Results: The absolute increase in unadjusted 5-year survival over 20 years was about 20% for stomach cancer, 14% for breast cancer and 5% for lung cancer. Lack of age adjustment would have caused proportionate under-estimation of these trends by 13–14% (stomach), 7% (breast), 14% (lung, men) and 4% (lung, women). Lack of adjustment for the trend toward earlier stage would have caused proportionate over-estimation (152%, men: 133%, women) of stomach cancer survival trends, which seemed more influenced by earlier diagnosis than more effective treatment. For breast cancer, the 31% over-estimation of trend from lack of stage adjustment suggests the impact of earlier diagnosis, while the improvement of survival after additional adjustment for age may be due to more effective treatment. Failure to adjust for stage led to a proportionate 21% under-estimation of lung cancer survival trends for men, and 4% over-estimation for women.
Conclusion: This study confirms the importance of adjusting for trends in age and stage distribution when evaluating time trends in cancer survival.
Key Words: Survival analysis • cancer registries • risk adjustment • early diagnosis • ageing
|
INTRODUCTION |
|---|
|
TOPAbstractINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
Five-year survival is an important index to evaluate the overall effectiveness of cancer diagnosis and treatment. An increase of 20% in 5-year cancer survival has been proposed as a possible objective for the Japanese government's ‘Strategy of Healthy Frontier’ between 2005 and 2015 (1). When the objective is finally agreed, it will be necessary to monitor progress towards the target figure. Progress should be evaluated with data from the seven population-based prefectural cancer registries that have already published survival data, which can be used as a baseline (2). The cancer population will change over the next decade, however, with an increase in the average age at diagnosis and a shift toward earlier stage at diagnosis, both of which will influence trends in overall survival. Stratification or adjustment for these variables will be necessary for valid comparison of cancer survival over time (3).
The aim of this study was to monitor time trends in 5-year survival while adjusting for the influence of trends in the distribution of stage and age at diagnosis. Previous studies using population-based cancer registry data have shown time trends in cancer survival adjusted either for age or for stage at diagnosis (4). In this study, we have adjusted simultaneously for both age and stage, and compared the impact of this joint adjustment with the impact of adjusting for each variable separately.
|
MATERIALS AND METHODS |
|---|
|
TOPAbstractINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
We have examined survival trends among residents of Osaka Prefecture who were diagnosed with cancer of the stomach, lung or breast (women only) during the 20-year period 1975–94 and registered in Osaka Cancer Registry. All patients were followed up for at least 5 years to determine their vital status.
Data Sources
Osaka Cancer Registry has collected population-based data on all residents of Osaka Prefecture diagnosed with a first primary malignancy since 1962. The prefectural population is approximately 8.8 million, of whom 2.6 million live in Osaka City. The registry has actively followed up each patient to ascertain their vital status (4). Active follow-up was not conducted for patients residing in Osaka City during the period 1975–92, however, so the analysis reported here was restricted to residents of Osaka Prefecture other than Osaka City itself. Data were retrieved from the registry database on patients resident outside Osaka City who were registered as newly diagnosed with a first primary malignancy of the stomach (ICD-10 code C16, 45 444 patients), lung (C33–34, 20 023 patients) or breast (C50, 12 727 women) during the 20-year period 1975–94. We were able to analyse data for 39 697 patients with stomach cancer, 17 400 with lung cancer and 11 988 women with breast cancer, for all of whom the stage at diagnosis was known. We estimated separately the survival for patients with unknown stage at diagnosis (9098 cases, 11.6% of total). A small proportion of cases were not followed up for 5 years after diagnosis (stomach, 1.7%; breast 3.0%; lung 0.6%). Sixteen patients (0.02%) diagnosed under 15 years of age were excluded from analysis.
Statistical Analysis
We calculated the absolute 5-year survival with the actuarial method for patients diagnosed in each of four calendar periods (1975–79, 1980–84, 1985–89, 1990–94). Survival estimates were also adjusted for age at diagnosis (five categories: 15–44, 45–54, 55–64, 65–74, 75+ years), stage (four categories: localized, involvement of regional lymph nodes, direct invasion of adjacent organs or tissue and distant metastasis) and age and stage combined (20 categories). Adjustment was made with a method proposed by Brenner et al. (5–10). Using this approach for age adjustment, for example, age-specific weights are first assigned to each patient in a given age group, and conventional survival analysis is then carried out with the ‘weighted individual data’, in which weights are applied to the contributions of each patient to the number of persons at risk of death. Thus if ri is the proportion of patients in the ith age group in a particular patient group (e.g. those diagnosed in 1980–84, 1985–89 and 1990–94), and si is the corresponding proportion in the standard population (patients diagnosed in 1975–79), then the weight assigned to each patient in the ith age group is si/ri. This method overcomes a problem that arises with the traditional approach (11) to direct standardization, when data are sparse in some or all of the categories for which we wish to adjust the survival estimate. Although our data set was large, data were still sparse in some of the 20 categories of age and stage combined, in that we extended the single-factor approach to deal with two factors.
Interpreting trends in survival as a guide to progress, without adjustment for stage or age, would be complicated by the combined effect on survival of trends in the distribution of age and stage among cancer patients diagnosed in successive calendar periods. We therefore defined the reference value for the trend in survival as the difference between the survival estimate for patients diagnosed in 1975–79 (baseline) and the survival estimate for patients diagnosed in 1990–94, after adjustment for changes in the distribution of age and stage since 1975–79. The impact of adjusting for stage or age was quantified by the proportionate difference from the reference trend obtained by adjusting only for age or for stage (see Appendix).
|
RESULTS |
|---|
|
TOPAbstractINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
The proportion of patients diagnosed with stomach cancer at localized stage increased substantially from about 25% in the late 1970s to over 40% in the early 1990s, even though the age distribution has shifted toward more elderly patients (Table 1). The age distribution of women with breast cancer also shifted slightly toward the elderly, and the stage distribution has shifted towards earlier stage at diagnosis: more than half the women with known stage diagnosed during the early 1990s had localized disease. The shift toward more elderly patients was more marked for lung cancer, but the stage distribution has not changed much since the early 1980s: more than 40% of patients still present with metastatic disease; the proportion was actually lower in the late 1970s.
|
Stomach
Five-year survival from stomach cancer increased rapidly in both sexes, from about 25% for patients diagnosed during 1975–79 to 45% for those diagnosed during 1990–94 (Fig. 1a,b). For men, the estimate of the trend in survival without adjustment (raw trend), a 19% increase, was hardly altered by adjustment for age (21% increase). By contrast, adjustment for stage substantially altered the estimate of trend in survival (only a 7% increase since 1975–79, and 8% increase when fully adjusted). Thus fully adjusted survival only increased from 26 to 34% and little change occurred after the early 1980s.
|
Compared with the fully adjusted trend in survival between 1975–79 and 1990–94, taken as the standard, the trend would have been under-estimated by a proportionate 13–14% if age adjustment had not been done (Iage, –0.139 for men and –0.134 for women, Table 2). More seriously, failure to adjust for the marked shift toward earlier stage at diagnosis would have led to a greater than two-fold over-estimation of the increase in survival (152.4% too high for men and 132.5% for women, corresponding to Istage 1.524, 1.325; Table 2).
|
Breast
Five-year survival from breast cancer also increased, from about 67% for women diagnosed during 1975–79 to the fully adjusted figure of 79% for women diagnosed during 1990–94 (Fig. 1c). The estimate of the fully adjusted trend (12% increase) was similar to the unadjusted raw trend (14% increase), with 15% and 11% increases after adjustment for age or stage, respectively. Little change in the fully adjusted survival occurred between 1985–89 and 1990–94.
The overall 12% increase in fully adjusted breast cancer survival between 1975–79 and 1990–94 would have been slightly under-estimated if age adjustment had not been done (a proportionate 6.6%, Iage, –0.066; Table 2). More importantly, lack of adjustment for the shift toward earlier stage at diagnosis would have led to over-estimation of the increase in survival by almost a third (31.9%, Istage, 0.319).
Lung
Five-year survival from lung cancer increased from 6.0% in men and 5.4% in women diagnosed 1975–79 to the fully adjusted figure of 14.1% in men and 14.6% in women for 1990–94, overall increases of 8.1 and 9.2%, respectively (Fig. 1d,e). These overall increases would have been under-estimated by a proportion of 13.9% for men and 3.5% for women if age adjustment had not been done (Iage, –0.139, –0.035; Table 2). Failure to adjust for stage would have caused proportionate under-estimation of the trend by 21.2% for men, and a slight over-estimation of 4.4% for women (Istage, –0.212, 0.044; Table 2).
The proportion of cases of unknown stage at diagnosis fell from 20% in 1975–79 to less than 10% in 1990–94: this trend was similar for all three cancers examined (data not shown). Five-year survival for unstaged cases was lower than the overall survival for all cases of known stage (data not shown): the difference was largest in stomach cancer (all staged cases 37%, unstaged 20%). For each cancer, however, the overall 5-year survival for all cases combined, both staged and unstaged, was not markedly different from survival for cases with known stage, about 2% lower for stomach cancer and less than 1% lower for cancers of the breast and lung: this is because only about 12% of tumours were of unknown stage (data not shown).
|
DISCUSSION |
|---|
|
TOPAbstractINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
Monitoring long-term trends in survival will often require adjustment for changes in the characteristics of patients diagnosed in successive calendar periods. Age and stage are perhaps the most important of these characteristics. When we sub-divide patient populations into many categories defined by such variables, sparse data may complicate the adjustment. Brenner's approach resolves this problem and it enabled us to quantify the impact of changes in the distribution of age and stage at diagnosis on observed survival trends in Osaka, by comparing survival trends adjusted by either age or stage alone with the fully adjusted trend.
For stomach cancer, both unadjusted and age-adjusted survivals rose continuously from 1975–79 to 1990–94, but stage-adjusted and fully adjusted survival levelled off in the 1980s. This seems to result from a trend to progressively earlier diagnosis. Awareness of the importance of early detection of stomach cancer in daily clinical practice and the spread of screening programmes, commonly provided by local communities or employers in Japan, has probably contributed to an increase in early detection (12,13). The trends we report here suggest that, after adjustment for a trend toward earlier diagnosis, treatment did not lead to much improvement in 5-year survival between the early 1980s and the early 1990s.
For women with breast cancer, 5-year survival increased substantially even after adjustment for age and stage. Earlier diagnosis and improved treatment would both seem to have contributed to the improvement. Mass screening with mammography for women aged 50 and over was only introduced in 2000 (40 years and over from 2004), but opportunistic mammography began to be widely used from the 1990s and it has been credited with earlier diagnosis of breast cancer (14). Additionally, breast-conserving surgery, combined with pre-operative or post-operative chemotherapy, and with radiotherapy, also became more widely used between the 1980s and the 1990s (15).
For lung cancer, stage-adjusted and fully adjusted 5-year survival increased more than the unadjusted trend. The higher proportion of cases diagnosed with distant metastases from 1980 onwards is probably due to improved diagnostic procedures. Some patients who would not have been diagnosed as metastatic in 1975–79 might have been so diagnosed with the techniques that became available later, especially after the introduction and widespread use of computerized tomography (16). This ‘stage migration’ is probably responsible for the increase in stage-specific and stage-adjusted survival (data not shown) (17).
In this study, we used the same age categories (15–44, 45–54, 55–64, 65–74 and 75+) for each cancer. Gondos et al. (9) have examined many methods to categorize age groups because the age distribution differs between cancers, e.g. prostate cancer patients are older. Further examination of how to categorize age in respect to the size of sample and the balance between age groups will be required.
The influence of stage migration needs to be borne in mind when we examine trends in fully adjusted survival. Stage migration may also affect time trends in stage-specific survival, as with lung cancer in this study.
The survival of patients with unknown stage at diagnosis is generally lower than that of patients with known stage, so their exclusion will tend to over-estimate survival in the population as a whole. Ideally, overall survival estimates used for public health evaluation of cancer control should include all cancer patients, even those for whom stage at diagnosis is unknown. The effect on overall survival estimates of excluding unstaged cases was not large in these data, however, since stage was known for a high proportion of cases.
This study suggests that adjustment for changes in the distribution of age and stage at diagnosis of cancer patients diagnosed in successive calendar periods may provide a very different picture of trends in survival than that obtained from unadjusted trends. It follows that Government policy to improve cancer survival—and the surveillance of progress toward any ‘target’ survival rate—should incorporate the impact of trends in the distribution of age and stage, in order to avoid spurious claims about success or failure of the policy.
|
APPENDIX |
|---|
|
TOPAbstractINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
Estimating the Impact on Survival Trends of Adjusting for Age and Stage
The overall proportionate change in 5-year survival (
) between patients diagnosed during 1975–79 and those diagnosed during 1990–94 was defined as follows, where S(t) is the 5-year survival rate at time t, with baseline time period t0 as 1975–79 and t1 as 1990–94.
|
| (1) |
|
| (2) |
The impact of adjusting the survival trend for stage Istage can be written as follows, and the values interpreted equivalently:
|
| (3) |
Conflict of interest statement
None declared.
|
Acknowledgment |
|---|
This study was supported by a Grant-in-Aid for Cancer Research from the Japanese Ministry of Health, Labour and Welfare: (14-2).
|
References |
|---|
|
TOPAbstractINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONAPPENDIXReferences |
|---|
1 Ministry of Health, Labour and Welfare. Commission of screening and health guidance for life-style related disease, the first meeting document, [in Japanese]. http://www.mhlw.go.jp/shingi/2005/07/s0725-7h.html.
2 Tsukuma H, Ajiki W, Ioka A, Oshima A. Survival of cancer patients diagnosed between 1993 and 1996: a collaborative study of population-based cancer registries in Japan. Jpn J Clin Oncol (2006) 36:602–7.
3 Parkin DM, Hakulinen T. Analysis of survival. In: Cancer Registration: Principles and Methods—Jensen OM, Parkin DM, MacLennan R, Muir CS, Skeet R, eds. (1991) Lyon: International Agency for Research on Cancer. 159–76. IARC Scientific Publications No. 95.
4 Ajiki W, Tsukuma H, Oshima A. Trends in cancer incidence and survival in Osaka. Gann Monogr Cancer Res (2004) 51:137–63.
5 Brenner H, Hakulinen T. On crude and age-adjusted relative survival rates. J Clin Epidemiol (2003) 56:1185–91.[CrossRef][ISI][Medline]
6 Brenner H, Arndt V, Geffeller O, Hakulinen H. An alternative approach to age adjustment of survival rates. Eur J Cancer (2004) 40:2317–22.[CrossRef][ISI][Medline]
7 Brenner H, Hakulinen T. Reduction in selective under-ascertainment bias in population-based estimates of cancer patient survival by age adjustment. Eur J Cancer (2005) 41:1788–93.[CrossRef][ISI][Medline]
8 Brenner H, Hakulinen T. Age adjustment of cancer survival rates: methods, point estimates and standard error. Br J Cancer (2005) 93:372–5.[CrossRef][ISI][Medline]
9 Gondos A, Parkin DM, Chokunonga E, Brenner H. Calculating age-adjusted cancer survival estimates when age-specific data are sparse: an empirical evaluation of various methods. Br J Cancer (2006) 94:450–4.[CrossRef][ISI][Medline]
10 Brenner H, Arndt V. Further enhanced monitoring of cancer patient survival by stage-adjusted period analysis. Cancer Epidemiol Biomarkers Prev (2005) 14:1917–21.
11 Verdecchia A, Capocaccia R, Santaquilani M, Hakulinen T. Methods of survival data analysis and presentation issues. In: IACR Scientific Publications No. 151—Berrino F, Sant M, Verdecchia A, Capocaccia R, Hakulinen T, Estève J, eds. (1999) Lyon: International Agency for Research on Cancer. 41–5. Survival of cancer patients in Europe: the EUROCARE-2 study.
12 Mizoue T, Yoshimura T, Tokui N, Hoshiyama Y, Yatsuya H, Sakata K, et al. Prospective study of screening for stomach cancer in Japan. Int J Cancer (2003) 106:103–7.[CrossRef][ISI][Medline]
13 Lee KJ, Inoue M, Otani T, Motoki I, Sasazuki S. Tsugane S for the JPHC Study Group. Gastric cancer screening and subsequent risk of gastric cancer: a large-scale population-based cohort study, with a 13-year follow-up in Japan. Int J Cancer (2006) 118:2315–21.[CrossRef][ISI][Medline]
14 Ohuchi N, Yoshida K, Kimura M, Ouchi A, Kamioki S, Shiiba K, et al. Improved detection rate of early breast cancer in mass screening combined with mammography. Jpn J Cancer Res (1993) 84:807–12.[CrossRef][ISI][Medline]
15 Ikeda T, Jinno H, Matsui A, Mitsui Y, Asaga S, Mutoh T, et al. Current status of breast conserving therapy in Japan. Biomed Pharmacother (2002) 56:182s–6s.[CrossRef]
16 Kaneko M, Kusumoto M, Kobayashi T, Moriyama N, Naruke T, Ohmatsu H, et al. Computed tomography screening for lung carcinoma in Japan. Cancer (2000) 89:2485–8.[CrossRef][ISI][Medline]
17 Feinstein AR, Sosin DM, Wells CK. The Will Rogers phenomenon: stage migration and new diagnostic techniques as a source of misleading statistics for survival in cancer. N Engl J Med (1985) 312:1604–8.[Abstract]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||