SEARCH

SEARCH BY CITATION

Keywords:

  • cancer registries;
  • prognosis;
  • survival

Abstract

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

International comparison of population-based cancer survival is a key component of monitoring progress against cancer. Its validity depends to an unknown degree on completeness of ascertainment of deaths in the cancer registries involved which may vary according to legal and administrative circumstances. The aim of this study was to assess the impact of incomplete registration of deaths through various mechanisms on the validity of long-term absolute and relative survival estimates. For that purpose, we simulated underascertainment of deaths through linkage failure of registry data with death certificates with probabilities between 0.1 and 5%, and underascertainment of deaths by unregistered annual emigration with probabilities between 0.05 and 2%, using data from the Finnish Cancer Registry. The expected impact on estimates of 5-, 10- and 15-year absolute and relative survival was assessed. We demonstrate that even modest levels of under-registration of deaths may lead to severe overestimation of long-term survival estimates, ranging from 0 to 31 percent units in the scenarios assessed. In general, relative survival is much more affected than absolute survival, and potential problems are much larger for relative survival estimates in older compared with younger patients. Potential overestimation strongly increases with length of follow-up, and this increase is particularly pronounced for under-registration of deaths because of unrecorded emigration. Every effort should be made in cancer registry based survival analyses to ascertain deaths with close to 100% completeness. When such completeness cannot be achieved, long-term relative survival estimates and their comparison across populations must be interpreted with much caution. © 2009 UICC

Providing estimates of long-term cancer survival on the population level is a key contribution of population-based cancer registries in monitoring progress against cancer. For example, the EUROCARE study has disclosed major differences and trends in cancer survival across European populations which had a profound impact on health policy planning on both the national and international level.1–4 Important prerequisites for obtaining valid survival estimates from population-based cancer registries are high quality and completeness of both incidence and mortality follow-up data.5 Several articles have addressed the potential impact on cancer survival estimates of less than perfect completeness of incidence data,6–8 but systematic quantitative investigation of less than perfect registration of deaths of cancer patients is sparse even though this very important issue has repeatedly come up in the interpretation of population-based cancer survival data.9, 10 The aim of this study was to assess and quantify the implications of incomplete registration of deaths on long-term survival estimates from population-based cancer registries.

Material and methods

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Our analyses are based on data from the nationwide Finnish Cancer Registry, which is internationally recognized for its high levels of data quality and completeness.11 We started from survival analyses based on actually observed mortality follow-up data which can be assumed to be complete due to highly efficient record linkage with national population registry and vital statistics data. In a next step, we modified the dataset to simulate various types of less than perfect registration of deaths and to study their effects on long-term survival estimates. The following two potential sources of underascertainment of deaths were assessed:a. imperfect linkage of death records with patient records in the registry,b. unregistered emigration of patients from the registration area.

Neither of these problems is likely to be of any relevance in Finland, but they can be of major concern in populations with less perfect record linkage with either vital statistics or population registries, in particular in local and regional registries where emigration from the registration area is much more common.

In our analyses, we included patients diagnosed with one of 20 common forms of cancer in Finland between 1985 and 1989 and followed with respect to mortality until the end of 2004. We excluded patients reported on the basis of an autopsy only (2.70%) or by death certificate only (0.88%), and patients with unknown month of diagnosis (0.12%). The cohort of patients diagnosed in 1985–1989 was selected, as this was the most recent cohort with complete follow-up for 15 years, the follow-up period covered in our analyses.

Specifically, we looked at 5-, 10-, and 15-year absolute and relative survival of these patients. Relative survival, which is commonly reported by population-based cancer registries, reflects the probability of surviving the cancer of interest rather than the total survival probability,12, 13 taking expected deaths in the absence of cancer into account. The expected numbers of deaths were derived from age, gender and calendar period specific mortality figures of the general population of Finland according to the Hakulinen method.14

To simulate imperfect record linkage, we recoded defined proportions of deceased patients as not having died and being censored (alive) until the end of follow-up (end of 2004), regardless of the time of their death. The numbers of patients at risk in subsequent years of follow-up were increased accordingly. To avoid random error, recoding was done in a deterministic rather than a stochastic manner (even though this led to noninteger numbers of registered deaths). Initially, 1% of deceased persons were reclassified in that way. In subsequent analyses, the proportion of missed links was varied between 0.1 and 5.0% of all deaths.

To simulate and assess the expected impact of unregistered emigration from the registration area, we assumed various levels of unregistered annual emigration probability (uae). Under the simplifying assumption that emigration is unrelated to prognosis, the probability of deaths being missed because of unregistered emigration was calculated as 1 − (1 − uae)t, where t denotes the survival time in years. Under the additional simplifying assumption that deaths are approximately equally spread within follow-up years, a proportion 1 − (1 − uae)y-0.5 of patients who died during each follow-up year y was assumed to be erroneously classified as surviving until the end of follow-up (end of 2004). Therefore, numbers of deaths in each year of follow-up and patients at risk in subsequent years of follow-up were adapted accordingly. Initially, unregistered annual emigration probability was set to 0.5%. This probability was varied between 0.05 and 2% in subsequent analyses. In addition to analyses for all ages combined, we carried out separate analyses for patients diagnosed below 75 years of age and older patients.

All analyses were carried out by the SAS statistical software package, using adapted versions of previously described macros for absolute and relative survival analysis.15 Although these macros were primarily developed for period analysis,15, 16 they can be used for cohort analyses (as done in this paper) as well.

Results

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

Numbers and mean age of patients diagnosed with the various forms of cancer in Finland in 1985–1989 who are included in this study are shown by cancer site in Table I. The minimum number of cases per site was 718 (cervical cancer). With more than 10,000 cases, breast and lung cancer were the most common forms of cancer in this sample. Mean age at diagnosis ranged from 52 years for cancers of the brain and of the thyroid gland to 73.5 years for prostate cancer.

Table I. Numbers and Mean Age of Patients Diagnosed With 20 Common Forms of Cancer in Finland in 1985–1989 and Included in the Survival Analyses
Cancer siteNumber of casesMean age
  • 1

    Only female cases were included.

Oral cavity1,99866.1
Esophagus1,03672.1
Stomach5,55770.0
Colon4,63968.9
Rectum3,17469.7
Liver2,28571.1
Pancreas3,17470.0
Lung10,64367.5
Breast111,64861.6
Cervix71862.8
Corpus2,51165.8
Ovary2,01762.7
Prostate6,21273.5
Kidney2,81365.5
Urinary bladder3,09469.8
Melanoma2,32957.0
Brain1,35852.1
Thyroid gland1,27052.2
Leukemia2,02364.4
Lymphoma1,85863.1

Using actually observed data, absolute 5-year survival ranged from 77.6% for thyroid cancer to 2.2% for pancreatic cancer (Table II). With the exception of thyroid cancer and melanoma, 15-year absolute survival was below 50% for all cancer patients. However, long-term relative survival was much higher, especially for cancers occurring at an older age, due to substantial overall mortality at those ages even in the general population. Failed record linkage for about 1% of deaths would have only a small impact on absolute survival estimates. Five-, 10- and 15-year survival would be expected to be overestimated by 0.2–1.0, 0.3–1.0 and 0.3–1.0 percent units for the various cancer sites. Relative survival would be more strongly affected, with expected overestimation of 0.2–1.3, 0.4–2.0 and 0.5–3.7 percent units for 5-, 10-, and 15-year relative survival. In general, overestimation is expected to increase with duration of follow-up and to be more pronounced for cancers occurring at relatively high ages, such as prostate cancer.

Table II. Five-, 10- and 15-Year Absolute and Relative Survival of Patients Diagnosed With Various Forms of Cancer in Finland in 1985–1989
Cancer siteAbsolute survivalRelative survival
Complete (y)Incomplete (y)Difference1 (y)Complete (y)Incomplete (y)Difference1 (y)
510155101551015510155101551015
  • Estimates obtained with complete registration and expected with 1% missed deaths in each year of follow-up due to imperfect record linkage.

  • 1

    Difference between estimates obtained with incomplete and complete ascertainment of deaths.

Oral cavity51.635.024.352.135.725.00.50.70.765.257.252.165.858.353.80.61.11.7
Esophagus6.34.22.67.25.13.60.90.91.08.68.27.69.910.010.41.31.82.8
Stomach16.411.88.417.312.69.30.90.80.921.921.621.823.023.324.11.11.72.3
Colon38.527.219.839.127.920.60.60.70.850.147.647.450.948.849.30.81.21.9
Rectum36.623.416.537.224.217.40.60.80.947.941.640.948.843.043.00.91.42.1
Liver4.93.32.15.94.33.11.01.01.06.66.15.67.97.98.11.31.82.5
Pancreas2.21.20.83.22.21.81.01.01.02.82.22.04.13.94.41.31.72.4
Lung8.74.62.59.65.63.50.91.01.011.07.95.912.29.58.21.21.62.3
Breast68.851.540.369.152.040.90.30.50.678.367.661.878.668.362.70.30.70.9
Cervix46.436.329.346.937.030.00.50.70.754.050.549.454.751.450.60.70.91.2
Corpus65.854.344.266.154.744.70.30.40.576.475.275.376.875.876.20.40.60.9
Ovary31.724.620.032.325.420.80.60.80.836.332.931.837.033.933.10.71.01.3
Prostate41.417.88.642.018.69.50.60.80.962.042.735.262.944.738.90.92.03.7
Kidney39.927.519.740.528.220.50.60.70.848.541.939.049.243.140.60.71.21.6
Urinary bladder49.731.720.650.232.421.40.50.70.867.160.256.567.861.558.70.71.32.2
Melanoma71.059.751.671.260.152.10.20.40.581.178.578.581.479.079.20.30.50.7
Brain27.520.116.028.220.916.80.70.80.830.224.622.031.025.623.20.81.01.2
Thyroid gland77.671.265.777.971.566.00.30.30.385.286.287.985.486.688.40.20.40.5
Leukemia29.915.49.130.616.210.00.70.80.937.424.518.438.325.820.30.91.31.9
Lymphoma36.023.417.436.724.118.20.70.70.843.734.932.444.436.134.00.71.21.6

Obviously, overestimation of survival increases with the degree of record linkage failure. This is demonstrated in Table III for 5-, 10- and 15-year relative survival. Overestimation would be expected to be negligible or very small (≤0.3 and ≤0.7 percent units for all of the 20 cancer sites assessed), even for 15-year relative survival, with linkage failure of 0.1 or 0.2% of deaths, respectively. On the other hand, expected overestimation would range up to 7.5 and 18.7 percent units for 15-year relative survival of patients with prostate cancer with linkage failure of 2 and 5%, respectively. For most other cancers and for shorter follow-up periods, expected overestimation would though be much smaller, even with these relatively severe forms of linkage problems.

Table III. Overestimation (In Percent Units) of 5-, 10- and 15-Year Relative Survival of Patients Diagnosed With Various Forms of Cancer in Finland in 1985–1989 Expected With Different Proportions of Missed Deaths Due to Imperfect Record Linkage
Cancer siteProportion of missed deaths due to imperfect record linkage
0.1%, y0.2%, y0.5%, y1.0%, y2%, y5%, y
510155101551015510155101551015
Oral cavity0.00.10.20.10.20.40.30.50.90.61.11.71.22.13.33.05.38.2
Esophagus0.10.20.30.30.30.60.70.91.41.31.82.82.63.75.76.59.414.2
Stomach0.10.20.20.20.40.40.60.81.11.11.72.32.23.34.75.68.111.8
Colon0.10.10.20.20.20.40.40.61.00.81.21.91.62.53.84.06.39.6
Rectum0.10.10.20.20.30.50.40.71.10.91.42.11.72.74.24.26.810.4
Liver0.10.20.20.20.40.50.60.91.31.31.82.52.53.65.16.38.912.7
Pancreas0.20.20.20.30.30.50.70.91.21.31.72.42.63.54.96.58.812.3
Lung0.20.10.20.30.30.40.60.81.11.21.62.32.43.24.65.98.111.5
Breast0.00.10.10.00.20.20.10.30.50.30.70.90.71.31.81.73.24.6
Cervix0.10.10.10.20.20.20.40.50.60.70.91.21.31.82.43.24.46.0
Corpus0.00.10.10.00.10.10.20.30.40.40.60.90.81.31.91.93.24.7
Ovary0.00.10.20.10.20.30.40.50.70.71.01.31.52.02.63.95.16.4
Prostate0.10.20.30.10.40.70.41.01.80.92.03.71.73.97.54.49.918.7
Kidney0.00.20.20.10.30.30.30.60.80.71.21.61.42.33.23.65.68.0
Urinary bladder0.10.10.20.20.30.40.40.71.10.71.32.21.42.64.43.46.510.9
Melanoma0.00.00.00.00.10.10.10.20.30.30.50.70.61.01.41.62.63.6
Brain0.10.10.10.20.20.20.40.50.60.81.01.21.62.02.34.04.95.8
Thyroid gland0.00.10.10.00.10.10.10.20.30.20.40.50.50.70.91.21.82.3
Leukemia0.10.10.20.20.20.40.50.71.00.91.31.91.82.73.74.46.79.3
Lymphoma0.00.10.20.10.20.40.30.60.80.71.21.61.52.33.13.85.77.8

Table IV shows expected estimates of 5-, 10- and 15-year absolute and relative survival in case of unregistered annual emigration of 0.5% of patients. Again, such unregistered emigration would only have a relatively small impact on absolute survival estimates. Five-, 10- and 15-year survival would be expected to be overestimated by 0.2–0.6, 0.4–1.4 and 0.4–2.0 percent units for the various cancer sites. Ranging between 0.1 and 0.9 percent units, expected overestimation would also be rather small for 5-year relative survival. However, for relative survival, expected overestimation would strongly increase with duration of follow-up, ranging from 0.5 to 3.5 percent units for 10-year relative survival, and from 0.9 to 8.2 percent units for 15-year relative survival. Again, overestimation would be expected to be strongest for prostate cancer, occurring at relative high ages and showing relatively high relative survival, whereas overestimation would be expected to be much lower (in absolute terms) for cancers occurring at relatively young ages such as thyroid cancer and cancers with very poor relative survival, such as pancreatic cancer.

Table IV. Five-, 10- and 15-Year Absolute and Relative Survival of Patients Diagnosed With Various Forms of Cancer in Finland in 1985–1989
Cancer siteAbsolute survivalRelative survival
Complete (y)Incomplete (y)Difference1 (y)Complete (y)Incomplete (y)Difference1 (y)
510155101551015510155101551015
  • Estimates obtained with complete registration and expected with unregistered emigration of 0.5% of patients per year

  • 1

    Difference between estimates obtained with incomplete and complete ascertainment of deaths.

Oral cavity51.635.024.352.036.125.90.41.11.665.257.252.165.758.955.70.51.73.6
Esophagus6.34.22.66.74.63.20.40.40.68.68.27.69.29.19.30.60.91.7
Stomach16.411.88.416.812.39.20.40.50.821.921.621.822.422.723.70.51.11.9
Colon38.527.219.838.928.021.10.40.81.350.147.647.450.749.050.40.61.43.0
Rectum36.623.416.537.124.417.90.51.01.447.941.640.948.643.444.40.71.83.5
Liver4.93.32.15.33.72.60.40.40.56.66.15.67.06.86.80.40.71.2
Pancreas2.21.20.82.51.61.20.30.40.42.82.22.03.32.82.90.50.60.9
Lung8.74.62.59.15.23.20.40.60.711.07.95.911.68.97.60.61.01.7
Breast68.851.540.369.252.541.90.41.01.678.367.661.878.768.964.30.41.32.5
Cervix46.436.329.346.837.130.50.40.81.254.050.549.454.651.651.50.61.12.1
Corpus65.854.344.266.155.045.50.30.71.376.475.275.376.776.277.50.31.02.2
Ovary31.724.620.032.125.321.00.40.71.036.332.931.836.833.933.40.51.01.6
Prostate41.417.88.642.019.210.60.61.42.062.042.735.262.946.243.40.93.58.2
Kidney39.927.519.740.328.421.10.40.91.448.541.939.049.043.341.70.51.42.7
Urinary bladder49.731.720.650.132.822.40.41.11.867.160.256.567.762.361.30.62.14.8
Melanoma71.059.751.671.360.452.80.30.71.281.178.578.581.479.480.30.30.91.8
Brain27.520.116.027.920.816.90.40.70.930.224.622.030.725.423.30.50.81.3
Thyroid gland77.671.265.777.871.666.40.20.40.785.286.287.985.386.788.90.10.51.0
Leukemia29.915.49.130.316.410.40.41.01.337.424.518.438.026.121.20.61.62.8
Lymphoma36.023.417.436.524.318.70.50.91.343.734.932.444.236.334.80.51.42.4

Clearly, expected overestimation would strongly depend on the extent of unregistered emigration. This is illustrated for relative survival in Table V. As long as the latter remains as low as 0.05% or lower per year, expected overestimation would remain mostly negligible even for 15-year relative survival. For unregistered emigration up to 0.2% per year, expected overestimation of 5-year relative survival would still remain below 0.4 percent units for all cancers, and expected overestimation of 10-year relative survival would still remain below 1 percent unit for all cancers but prostate cancer. On the other hand, quite extreme overestimation between 3.7 and 31.3 percent units would be expected for 15-year relative survival in case of unregistered annual emigration of 2% of cases. At these levels of unregistered emigration, expected overestimation would range from 0.6 to 3.6 percent units for 5-year relative survival, and from 1.8 to 13.5 percent units for 10-year relative survival.

Table V. Overestimation (In Percent Units) of 5-, 10- and 15-Year Relative Survival of Patients Diagnosed With Various Forms of Cancer in Finland in 1985–1989 Expected With Different Proportions of Unregistered Patient Emigrations Per Year
Cancer siteUnregistered annual emigration of patients
0.05%, y0.1%, y0.2%, y0.5%, y1%, y2%, y
510155101551015510155101551015
Oral cavity0.00.20.40.10.30.80.20.71.50.51.73.61.13.37.12.16.513.7
Esophagus0.10.10.20.10.20.30.20.40.70.60.91.71.21.93.42.33.86.7
Stomach0.10.10.20.10.20.40.20.40.80.51.11.91.12.13.92.14.17.6
Colon0.10.10.30.10.30.60.20.51.20.61.43.01.12.86.02.25.511.5
Rectum0.10.20.40.20.40.80.30.71.50.71.83.51.43.56.92.86.913.4
Liver0.00.10.10.10.20.20.20.30.50.40.71.20.91.52.41.82.94.8
Pancreas0.10.10.10.10.10.20.20.20.40.50.60.90.91.31.91.82.53.8
Lung0.10.10.10.20.20.30.30.40.70.61.01.71.22.03.42.44.16.7
Breast0.00.20.30.00.30.50.10.51.00.41.32.50.82.65.01.65.09.6
Cervix0.10.10.20.20.20.40.30.50.80.61.12.11.12.24.02.24.47.8
Corpus0.00.10.20.00.20.40.10.40.90.31.02.20.61.94.31.33.78.3
Ovary0.00.10.20.10.20.40.20.40.70.51.01.61.12.03.22.13.86.2
Prostate0.10.40.80.20.71.60.31.43.30.93.58.21.86.916.13.613.531.3
Kidney0.00.20.30.10.30.60.20.61.10.51.42.71.02.65.22.05.110.0
Urinary bladder0.10.20.50.10.41.00.30.92.00.62.14.81.34.19.52.58.118.3
Melanoma0.00.10.10.00.10.30.10.30.70.30.91.80.71.83.51.33.66.8
Brain0.10.10.10.10.20.30.20.30.50.50.81.30.91.62.51.83.24.8
Thyroid gland0.00.10.10.00.10.20.00.20.40.10.51.00.31.01.90.61.83.7
Leukemia0.10.10.30.20.30.60.30.61.20.61.62.81.23.15.52.46.110.7
Lymphoma0.00.20.30.10.30.50.20.61.00.51.42.41.02.74.72.15.39.0

Table VI demonstrates the variation of expected overestimation of relative survival because of unregistered emigration between age groups. With 0.5% unregistered annual emigration, expected overestimation of 5-, 10- and 15-year relative survival would reach maximum levels of 0.7, 2.4 and 5.2 percent units for the 20 common forms of cancer included in this analysis among patients below age 75. Much higher levels of overestimation, especially for very long-term relative survival, would be expected for age group 75+, ranging up to 1.3, 6.9 and 29.1 percent units for 5-, 10- and 15-year relative survival, respectively. The results shown in Table VI thereby indicate that the most severe forms of bias are essentially confined to long-term relative survival in older patients. However, even within this group of patients, major variation in the expected bias was observed. In general, much more severe overestimation of relative survival (in terms of differences from true relative survival) is expected for cancers with higher relative survival than for highly fatal cancers.

Table VI. Five-, 10- and 15-Year Relative Survival of Patients Diagnosed with Various Forms of Cancer in Finland in 1985–1989 in Two Major Age Groups
Cancer site<75 y at diagnosis≥75 y at diagnosis
Complete (y)Incomplete (y)Difference1 (y)Complete (y)Incomplete (y)Difference1 (y)
510155101551015510155101551015
  • Estimates obtained with complete registration and expected with unregistered emigration of 0.5% of patients per year.

  • 1

    Difference between estimates obtained with incomplete and complete ascertainment of deaths.

Oral cavity66.159.254.166.560.456.60.41.22.563.147.836.564.153.258.31.05.421.8
Esophagus11.59.98.912.010.710.10.50.81.23.73.20.04.44.95.50.71.75.5
Stomach25.523.722.926.024.524.30.50.81.414.414.716.215.116.924.20.72.28.0
Colon53.349.549.253.750.551.10.41.01.943.642.440.644.445.955.30.83.514.7
Rectum50.743.342.751.344.645.00.61.32.341.636.532.142.740.648.41.14.116.3
Liver8.06.46.18.47.07.00.40.60.93.85.42.34.46.87.00.61.44.7
Pancreas3.12.32.03.52.82.70.40.50.72.21.91.82.73.15.70.51.23.9
Lung12.58.66.213.09.57.70.50.91.54.62.40.95.44.57.50.82.16.6
Breast79.768.763.080.069.664.80.30.91.872.362.552.873.267.272.00.94.719.2
Cervix59.554.451.559.955.353.10.40.91.633.325.726.134.328.937.31.03.211.2
Corpus81.979.177.482.179.879.10.20.71.751.647.352.452.450.866.00.83.513.6
Ovary39.735.233.240.236.134.50.50.91.319.113.911.019.816.118.50.72.27.5
Prostate62.543.636.263.246.041.40.72.45.261.942.135.063.249.064.11.36.929.1
Kidney51.043.139.551.544.241.70.51.12.237.234.738.838.138.051.60.93.312.8
Urinary bladder71.663.659.772.165.062.90.51.43.257.148.935.958.254.159.01.15.223.1
Melanoma82.679.679.482.980.280.70.30.61.371.470.475.972.575.497.31.15.021.4
Brain32.325.522.432.826.323.60.50.81.21.22.57.01.83.610.20.61.13.2
Thyroid gland90.089.789.490.190.090.10.10.30.738.226.351.238.929.962.70.73.611.5
Leukemia40.226.319.440.827.621.60.61.32.228.514.79.029.418.221.20.93.512.2
Lymphoma48.937.434.249.338.636.10.41.21.922.918.410.123.721.220.70.82.810.6

Discussion

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

This paper provides a quantitative assessment of the potential impact of imperfect mortality follow-up on measures of long-term survival commonly derived from population-based cancer registry data. We demonstrate that even modest levels of under-registration of deaths may lead to severe overestimation of long-term survival estimates. In general, relative survival is much more affected than absolute survival, and potential problems are much larger for relative survival estimates in older compared with younger patients. These patterns are due to the fact that the expected changes in absolute survival are magnified through division by the expected survival proportions (derived from population life tables), the latter being much lower in older patients. Potential overestimation strongly increases with length of follow-up, and this increase is particularly pronounced for under-registration of deaths because of unrecorded emigration.

Virtually complete registration of deaths can be achieved in countries, where reliable and comprehensive record linkage of registry records with both population registry and vital statistics data is possible which is the case in the North European countries including Finland. Here, very reliable record linkage is possible because of the use of unique personal identification numbers, both in the national cancer registry and the national population registry. Deaths of cancer patients occurring after emigration from the country might still be missed. However, whereas migration within the country is quite common, emigration across country borders of patients with cancer is still relatively rare. Furthermore, such emigration will not affect survival estimates, as long as the date of emigration is reliably recorded (which is ensured by comprehensive population registries) and observations are censored at that date, and provided that emigration is unrelated to prognosis.

Problems such as those assessed in this manuscript can, however, easily occur under less favourable circumstances. For example, many cancer registries have to rely on combinations of multiple personal identifiers, such as first and last name, address, date of birth, etc. for record linkage, making the latter vulnerable by multiple sources, such as typing errors, changes of names or addresses, etc. The situation may even be worse in countries, where no unique person numbers exist, and access and use of personal identifiers for record linkage is strongly limited by overly restrictive data protection rules. Unregistered emigration may be a particularly salient problem in regional registries exclusively relying on record linkage with notifications of deaths in the registration area for mortality follow-up, without possibility to reliably and comprehensively identify emigration from the registration area through population registries. In such situations, it may often be very tempting and sometimes the only choice to assume patients to be alive unless their death was specifically reported through a death certificate. However, our analyses clearly show, that this assumption may lead to severe overestimation of long-term relative survival estimates, especially for very long-term follow-up periods and for older age groups, even at levels of unregistered emigration that might appear to be small enough to be negligible on first view.

Although the issues addressed in this manuscript may be relevant to some degree for any population based cancer survival analysis, they may be of particular concern for national and international comparative survival analyses. In such analyses, different levels of completeness of recording of deaths in the registries involved might lead to spurious differences in survival, favouring those registries with less perfect mortality follow-up. In such comparative analyses, the problem of comparability may even be aggravated by commonly employed methods of age adjustment intended to make results comparable between cancer populations with different age structure. Conventional direct age adjustment provides a weighted average of survival across various age groups. The age specific weights are kept constant, regardless of the length of follow-up, and reflect the age distribution of some cancer population at the time of diagnosis, such as the world cancer population,17 or the international cancer survival standards.18 However, with increasing length of follow-up, the age distribution of surviving patients is strongly shifted to younger ages, which is implicitly taken into account in analysis of crude relative survival, but not in conventional age adjustment of relative survival.19 The greater weight given to the oldest age groups with increasing length of follow-up will make the age adjusted estimates of long-term relative survival even more susceptible to bias because of incomplete recording of deaths.

This manuscript addressed only two, albeit important types of problems of recording deaths of cancer patients in cancer registries. Other problems might also be relevant in particular settings. For example, the occurrence of false positive links of deaths with registered patients may also be possible in settings where reliable unique personal identifiers are not available.20 Such false positive links may partly balance overestimation of survival due to false negative links. On the other hand, incomplete ascertainment of deaths for other reasons, especially in situations with lack of comprehensive vital statistics (encountered in many developing countries), may further aggravate potential overestimation of survival.

In summary, complete ascertainment of deaths is of utmost importance for valid estimation of cancer survival, especially long-term relative survival. Every effort should be made in cancer registration to ascertain deaths with close to 100% completeness. Legal and administrative regulations need to ensure that this goal can be reliably achieved. Regular checks for and focused active follow-up of long-term “survivors” who seem to survive for exceptionally long time given their age, cancer site or stage may be an additional approach to enhance completeness of death ascertainment. When close to 100% completeness cannot be achieved, estimates of long-term relative survival must be interpreted with much caution.

Acknowledgements

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References

The work of Hermann Brenner was partly supported by a grant from the German Cancer Foundation (Deutsche Krebshilfe, Project No. 108257). Timo Hakulinen's work was supported by grants from the Academy of Finland and the Cancer Society of Finland.

References

  1. Top of page
  2. Abstract
  3. Material and methods
  4. Results
  5. Discussion
  6. Acknowledgements
  7. References
  • 1
    BerrinoF,CapocacciaR,EstèveJ,GattaG,HakulinenT,MicheliA,SantM,VerdecchiaA, eds. Survival of cancer patients in Europe. The EUROCARE-2 study. IARC Scientific Publications No. 151. Lyon: IARC, 1999.
  • 2
    BerrinoF,CapocacciaR,ColemanMP,EstèveJ,GattaG,HakulinenT,MicheliA,SantM,VerdecchiaA, eds. Survival of cancer patients in Europe. The EUROCARE-3 Study. Ann Oncol 2003; 14 ( Suppl 5): v9155.
  • 3
    Berrino F,De Angelis R,Sant M,Rosso S,Lasota MB,Coebergh JW,Sanquilani M,the EUROCARE Working Group. Survival for eight major cancers and all cancers combined for European adults diagnosed in 1995-1999: results of the EUROCARE-4 study. Lancet Oncol 2007; 8: 77383.
  • 4
    Verdecchia A,Francisci S,Brenner H,Gatta G,Micheli A,Mangone L,Kunkler I,the EUROCARE-4 Working Group. Recent cancer survival in Europe: a 2000-02 period analysis. Lancet Oncol 2007; 8: 78496.
  • 5
    Capocaccia R,Gatta G,Roazzi P,Carrani E,Santaquilani M,De Angelis R,Tavilla A,EUROCARE Working Group. The EUROCARE-3 database: methodology of data collection, standardisation, quality control and statistical analysis. Ann Oncol 2003; 14 ( Suppl 5): v1427.
  • 6
    Brenner H,Hakulinen T. Population-based monitoring of cancer patient survival in situations with imperfect completeness of cancer registration. Br J Cancer 2005; 92: 5769.
  • 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: 178893.
  • 8
    Robinson D,Sankila R,Hakulinen T,Moller H. Interpreting international comparisons of cancer survival: the effects of incomplete registration and the presence of death certificate only cases on survival estimates. Eur J Cancer 2007; 43: 90913.
  • 9
    Sriamporn S,Swaminathan R,Parkin DM,Kamsa-ard S,Hakama M. Loss-adjusted survival of cervix cancer in Khon Kaen, Northeast Thailand. Br J Cancer 2004; 91: 10610.
  • 10
    Swaminathan R,Rama R,Shanta V. Lack of active follow-up of cancer patients in Chennai, India: implications for population-based survival estimates. Bull World Health Org 2008; 86: 50915.
  • 11
    Teppo L,Pukkala E,Lehtonen M. Data quality and quality control of a population-based cancer registry. Experience in Finland. Acta Oncol 1994; 33: 3659.
  • 12
    Ederer F,Axtell LM,Cutler SJ. The relative survival rate: a statistical methodology. Natl Cancer Inst Monogr 1961; 6: 10121.
  • 13
    Henson DE,Ries LA. The relative survival rate. Cancer 1995; 76: 16878.
  • 14
    Hakulinen T. Cancer survival corrected for heterogeneity in patient withdrawal. Biometrics 1982; 38: 93342.
  • 15
    Brenner H,Gefeller O,Hakulinen T. Period analysis for ‘up-to-date’ cancer survival data: theory, empirical evaluation, computational realisation and applications. Eur J Cancer 2004; 40: 32635.
  • 16
    Brenner H,Hakulinen T. Up-to-date long-term survival curves of patients with cancer by period analysis. J Clin Oncol 2002; 20: 82632.
  • 17
    Sankaranarayanan R,Black RJ,Parkin DM. Cancer survival in developing countries. IARC Scientific Publications No. 145. Lyon: IARC, 1998.
  • 18
    Corazziari I,Quinn M,Capocaccia R. Standard cancer patient population for age standardising survival ratios. Eur J Cancer 2004; 40: 230716.
  • 19
    Brenner H,Hakulinen T. On crude and age adjusted relative survival rates. J Clin Epidemiol 2003; 56: 118591.
  • 20
    Brenner H,Schmidtmann I,Stegmaier C. Effects of record linkage errors on registry-based follow-up studies. Stat Med 1997; 16: 263343.