5. RESULTS
5.2 Data quality indices for population-based survival study
The comparison of descriptive statistics on data quality indices of population-based cancer survival for lung and breast cancers between selected registries classified on methods of cancer registration and follow up are given in Table 5 (SURVCAN).
Table 5: Data quality indices: Frequency of excluded cases expressed as proportion of death certificate only and no follow up cases for lung and breast cancers in selected registries from low or medium resource countries separately for passive and active methods of cancer registration and follow up, 1990-2001*
(SURVCAN database)
Lung Breast
Country/Registry Total registered
DCO% NFU% Included
%
Total registered
DCO% NFU% Included
% Passive methods of
registration and follow up
China, Shanghai 14,113 0.0 0.1 99.9 5,184 0.0 0.0 100.0
Costa Rica DNA DNA DNA DNA 2,854 2.5 11.2 86.3
Korea, Seoul 10,294 12.7 4.9 82.4 5,907 3.3 6.0 90.7
Thailand, Lampang 3,278 7.8 0.3 91.9 842 1.4 0.0 98.6
Range 0-13 0-5 82-100 0-3 0-11 86-100
Active methods of registration and follow up
China, Qidong 3,303 0.3 0.4 99.3 669 0.3 2.4 97.3
India, Barshi 48 2.1 0.0 97.9 124 0.0 0.0 100.0
India, Mumbai 3,995 13.0 0.6 86.4 7751 5.2 0.7 94.1
Thailand, Songkhla 850 5.1 16.1 78.8 665 1.2 13.3 84.5
Range 0-13 0-16 79-99 0-5 0-13 85-100
DCO: Death certificate only; NFU: No follow up; DNA: Data not available
* Period varies for individual registries
The figures for frequency of lung cancer cases included for survival analysis ranged between 82-100% for selected registries in Asia and Central America, employing passive methods and 79-99% for selected registries in Asia undertaking active methods (Table 5).
The corresponding figures for breast cancer were 86-100% and 85-100% respectively. The range of DCOs for lung cancer was 0-13% in both groups of registries; the range for no follow up cases was 0-5% in registries with passive methods and 0-16% in registries with active methods for lung cancer. The range of DCOs for breast cancer was 0-3% in registries with passive methods and 0-5% in registries with active methods; the corresponding figures for no follow up cases were 0-11% and 0-13% respectively. The comparison of frequency of cases included for survival analysis out of total incident cases between the two groups of registries that pursued passive or active methods of case registration and follow up revealed minimal variation for lung and breast cancers (Table 5). This augurs well for the conduct of a population-based cancer survival study.
However, the frequency of cancer cases excluded from survival study (owing to being DCOs or no follow up (NFU) with zero survival time) and the frequency of cases with incomplete or loss to follow up (LFU) among cases included in survival study both have to be considered in unison to evaluate data quality in population-based survival study. There may be instances when one is minimal while the other is not. The following scenarios from real data present the different problems encountered pertaining to data quality indices and reiterate their possible impact on the population-based cancer survival.
Table 6: Frequency of cases registered as DCO or with lack of complete follow up for common cancers in Cuba, 1994-1995 followed through 1999*
Excluded from survival study
Incomplete follow up: % lost to follow up- years from diagnosis
Cancer/site Total
registered DCO% NFU%
Included in analysis
%
Complete follow up
% <1 1-3 3-5 >5
Tongue 314 25.5 0.9 73.6 95.2 3.0 0.9 0.9 0.0
Mouth 355 25.9 0.9 73.2 93.4 3.1 1.9 1.2 0.4
Tonsil 82 32.9 1.2 65.9 98.1 0.0 0.0 1.9 0.0
Oropharynx 60 20.0 0.0 80.0 97.9 2.1 0.0 0.0 0.0
Colon 2491 49.7 0.2 50.1 99.2 0.2 0.2 0.3 0.1
Rectum 790 29.1 0.5 70.4 98.0 1.1 0.4 0.5 0.0
Anus 106 8.5 2.8 88.7 98.9 1.1 0.0 0.0 0.0
Larynx 1165 30.7 0.6 68.7 96.3 1.5 0.9 1.0 0.3
Breast 2929 25.6 0.3 74.1 97.0 0.9 0.3 1.4 0.4
Cervix uteri 1450 15.4 0.4 84.2 94.4 2.2 1.2 1.7 0.5
Urinary bladder 1182 29.5 1.2 69.3 97.4 1.2 0.0 1.0 0.4
Hodgkin lymphoma 320 40.3 1.6 58.1 97.4 1.6 0.5 0.5 0.0
Non Hodgkin lymphoma 771 39.7 0.1 60.2 97.7 0.6 0.6 1.1 0.0
DCO: Death certificate only; NFU: No follow up or lost to follow up with zero survival time;
* SURVCAN database
Table 6 shows the frequencies of cases pertaining to all the data quality indices of population-based survival for selected cancers in Cuba during 1994-1995 and followed through 1999 (SURVCAN). Registration of cancer cases was carried out entirely by passive method. The frequency of cases registered as DCO ranges between 9-50% for different cancers. The very high figure may be the result of lack of active method of tracing back the cancer cases, first identified through a death certificate, to hospitals or to other sources of registration. On the other hand, the frequency of cases with zero survival time and vital status unknown were negligible ranging between 0-3%. In total, the frequency of cases included for survival analysis ranged between 50-89%. This is quite low and may or may not be a random sample or representative of the total incident cases. On the other hand, among the cases included for survival analysis, the complete follow up was achieved in 94-99%, which is adequate (Table 6). Follow up for vital status information was carried out predominantly by passive methods with minimal active component. Overall, despite good follow up, the resulting survival may not reflect the average outcome of respective cancers in the region owing to high degree of exclusion from the survival study thereby indicating high selection of cases.
The other kind of problem pertaining to inadequate follow up usually encountered in real data from low or medium resource countries is described in Table 7 (SURVCAN).
Table 7: Frequency of cases registered as DCO or with lack of complete follow up for common cancers in Khon Kaen, Thailand, 1993-1997 followed through 2000*
Excluded from survival study
Incomplete follow up: % lost to follow up- years from diagnosis
Cancer/site Total
registered DCO% NFU%
Included in analysis
%
Complete follow up
% <1 1-3 3-5 >5
Lip 88 1.1 9.1 89.8 67.1 19.0 5.1 6.3 2.5
Tongue 57 0.0 5.3 94.7 77.8 20.3 1.9 0.0 0.0
Mouth 120 3.3 7.5 89.2 73.8 19.6 2.8 1.9 1.9
Nasopharynx 123 0.0 4.1 95.9 80.5 12.8 4.2 0.8 1.7
Colon 258 1.6 4.7 93.8 79.8 6.6 6.2 5.4 2.0
Rectum 143 0.0 2.8 97.2 80.6 10.1 6.5 0.7 2.1
Larynx 38 0.0 5.3 94.7 80.6 19.4 0.0 0.0 0.0
Breast 446 1.1 5.2 93.7 41.4 24.4 22.2 7.7 4.3
Ovary 230 0.9 9.6 89.6 47.1 28.6 12.6 8.7 3.0
Urinary bladder 114 0.0 9.6 90.4 41.7 35.9 10.7 8.7 3.0
Hodgkin lymphoma 31 0.0 6.5 93.5 37.9 27.6 17.2 6.9 10.4
Non Hodgkin lymphoma 191 0.0 12.0 88.0 53.0 28.0 10.1 6.0 2.9 DCO: Death certificate only; NFU: No follow up or lost to follow up with zero survival time;
* SURVCAN database
Table 7 describes both the two data quality indices pertaining to cancer registration and follow up as observed in survival data from Khon Kaen, Thailand, in 1993-1997 and followed through 2000. Both cancer registration and follow up were carried out predominantly by active methods. The frequency of cases registered as DCO was minimal ranging between 0-3% for different cancers. This possibly is due to trace-back procedures by active methods adopted by the registry. However, the frequency of cases excluded from survival study with zero survival time and vital status unknown was higher ranging between 3-12%. In total, the frequency of cases included for survival analysis ranged between 88-97%, which is satisfactory. Among the cases included for survival analysis, the complete follow up was achieved only in 38-81% for different cancers. Overall, despite minimal exclusions from the study, the resulting absolute survival is not likely to reflect the average outcome of respective cancers in the region owing to the very high degree of incomplete or losses to follow up (19-62%), especially due to high losses to follow up within the first year from diagnosis (7-36%). These data quality indices, with the exception of DCOs, are applicable to hospital-based survival studies also.