Abstract

Introduction

Materials and Methods

Results

Discussion

Conclusions

Data Availability

Disclosure

Conflicts of Interest

Acknowledgments

Supplementary Materials

References

Copyright

Related articles

Abstract

     Background. Estimation of survival requires follow-up of patients from diagnosis until death ensuring complete and good quality data. Many population-based cancer registries in low- and middle-income countries have difficulties linking registry data with regional or national vital statistics, increasing the chances of cases lost to follow-up. The impact of lost to follow-up cases on survival estimates from small population-based cancer registries (<500 cases) has been understudied, and bias could be larger than in larger registries. Methods. We simulated scenarios based on idealized real data from three population-based cancer registries to assess the impact of loss to follow-up on 1-5-year overall and net survival for stomach, colon, and thyroid cancers—cancer types with very different prognosis. Multiple scenarios with varying of lost to follow-up proportions (1-20%) and sample sizes of (100-500 cases) were carried out. We investigated the impact of excluding versus censoring lost to follow-up cases; punctual and bootstrap confidence intervals for the average bias are presented. Results. Censoring of lost to follow-up cases lead to overestimation of the overall survival, this effect was strongest for cancers with a poor prognosis and increased with follow-up time and higher proportion of lost to follow-up cases; these effects were slightly larger for net survival than overall survival. Excluding cases lost to follow-up did not generate a bias on survival estimates on average, but in individual cases, there were under- and overestimating survival. For gastric, colon, and thyroid cancer, relative bias on 5-year cancer survival with 1% of lost to follow-up varied between 6% and 125%, 2% and 40%, and 0.1% and 1.0%, respectively. Conclusion. Estimation of cancer survival from small population-based registries must be interpreted with caution: even small proportions of censoring, or excluding lost to follow-up cases can inflate survival, making it hard to interpret comparison across regions or countries.

Introduction

     Population-based cancer registries (PBCR) are the gold standard source of population-based incidence and survival statistics [1, 2]. Estimation of survival requires follow-up of a cohort of cancer patients until their moment of death or study closure, ensuring complete follow-up and good quality data. Many PBCR, contrary to the situation of clinical trials, do not perform routine follow-up of all patients when they perform survival analyses. Usually, they cross-link information from civil registration and vital statistics of their population to be informed regarding date (and cause) of death. If the completeness or quality of the vital statistics is suboptimal, deceased patients may not be reported and therefore considered erroneously being alive at the date of administrative follow-up. The quality of the death-registry is relatively complete in Colombia [3]; therefore, absence of a death certificate is usually considered as an indicator to consider the patient alive. However, in Colombia and low- and middle-income countries (LMIC), this cross-linking between registries and vital statistics is difficult and often only possible for the regional vital statistics data—if a patient died in another city, the registries would not be notified. Sometimes strict personal data protection policies [4] inhibit this linkage altogether, and registries have to rely on notes made by registrars when coincidentally encountering patients already in the registry.

Results and analyses

     Records of 3,455 incident cancer patients were collected from the three PBCR, 1,528 (44.2%) for stomach, 902 (26.1%) for colon, and 1,025 (29.7%) for thyroid gland cancer. After applying the exclusion criteria, 2,924 cases remained: 1,188 for stomach, 759 for colon, and 977 for thyroid gland cancer (details for reasons for exclusion in table S1). The proportion of excluded tumors was 22.3% for stomach, 15.9% for colon, and 4.7% for thyroid gland. The mean age for stomach cancer cases was 63.8 years (standard deviation (SD): 15.0), for colon 64.8 years (SD 14.1), and for thyroid gland 48.4 years (SD 14.3). The proportion of women was 41.5, 60.5, and 86.4% for stomach, colon, and thyroid gland, respectively.

     Table 1 shows overall one-, three-, and 5-year all-cause and net survival for the complete case analysis of the three cancer types (reference scenario). For stomach cancer, one-year overall and net survival were 0.39 and 0.40; corresponding figures were 0.66 and 0.67 for colorectal cancer and 0.96 and 0.96 for thyroid cancer. Five-year overall and net survival were 0.19 and 0.21 for stomach cancer, 0.44 and 0.50 for colon cancer, and 0.92 and 0.95 for thyroid gland cancers, respectively. Net survival was always slightly higher than overall survival.