**Regarding articles about Cali Cancer Registry I**

*Dear editor:* With great interest we read the series of papers produced by the population-based cancer registry of Cali in Colombia on the situation in breast, colorectal and prostate cancer in this country^{1-3}

The Cali Cancer Registry has been of great value for cancer information and resource planning in Colombia for more than 50 years now. Besides providing valuable information on cancer incidence in an unbiased population, it has relatively recently also began to provide estimates of cancer survival on a population-basis. This is extremely important as knowing the prognosis of cancer patients, particularly in different population groups, gives us important information on the state of cancer diagnosis and treatment in a population, as it is influenced by stage at diagnosis (in its turn potentially influenced by early detection and timely access to healthcare facilities) and access to and quality of care.

However, as always, the survival statistics presented in the previously mentioned papers should, in our opinion, be interpreted with caution, particularly when comparing two different time periods. In all three referenced manuscripts, the 1, 3 and 5 year survival is reported for the period 1995-2004, with a follow-up period until December 31^{st}, 2006 for the papers on breast and colorectal cancer (for the prostate cancer paper the end of follow-up is not described). Although it is unclear why the follow-up was not extended to a later period as crosses were made with the mortality databases until 2010 and the SISBEN database of 2008, this end of follow-up until 2006 already means that for part of the population included in the survival analyses, 5-year survival information is not available, since follow-up for a person diagnosed in 2004 could have maximally been 3 years. Therefore, 5 year survival estimates are based on persons diagnosed longer ago, which is important to mention

More importantly, on colorectal cancer, the authors divided the observation period into two (1995-1999 and 2000-2004) and conclude that the 5-year relative survival improved significantly in men and women from Cali.^{1} Indeed, the estimates show statistically significant improvements, for example from 28 to 56% in males under 50 years of age. This is a doubling in survival time, and if it were true, would be very good news. However, these calculations are based on one cohort with complete follow-up (the 1995-1999 cohort) and one cohort in which only the patients diagnosed in 2000 and 2001 have reached 5 years follow-up. Patients diagnosed in the subsequent years were presumably censored at December 31^{st} 2006. Incompleteness of follow-up is known to generally cause inflated survival estimates,^{4} and this artificial fact may be causing part of the very large increases in survival. This seems likely to have happened, as we see smaller, although still large improvements occurring for the 1 year relative survival rates, and intermediate improvements in the 3 year survival rates presented in this paper. Indeed, the 2000-2004 cohort does not have complete follow-up at three years either. As the authors do not report to have used period-based relative survival models,^{5} we presume the presented results are the traditional cohort-based ones and therefore the second one is incomplete in follow-up.

Of course, real increases in survival may have occurred, but having such a large increase within just a few years on a population-level should be carefully evaluated. Changes in follow-up practices, coding, etc., may also be underlying such changes.

]]> In addition there is an error in the multivariate survival analysis methodology causing difficulty in its interpretation: Cox proportional hazards models were constructed to perform analyses investigating the effects of age, stage, socioeconomic status and period of diagnosis. The most important underlying assumption of Cox proportional hazards modeling is that the hazards are proportional.^{6}The authors performed a statistical test to evaluate if this assumption applied to their data, and this test clearly showed that in certain subgroups, the proportional hazards assumption was violated. Whereas this impedes continuing with conventional Cox proportional hazards modeling and calls for performing either stratified analyses or a time-dependent exposure models,

^{6}the authors ignored the violation and presented their results without consideration on it. Depending on the exact underlying data, this may result in invalid conclusions and therefore we feel that the results presented on the hazards of dying for the covariates should not interpreted at all.

We consider the point highly relevant, as the idea of performing multivariate analyses on survival of cancer in Colombia in a population-based setting is of major interest, particularly referring to socioeconomic differences.

We hope that the registry will have the opportunity to clarify some of the points we have presented as well as to improve statistical analysis in future studies.

**Esther de Vries , PhD,^{(1)} Raul Murillo, MD, MPH.^{(2)}**

*(1) Grupo Vigilancia Epidemiológica del Cáncer, Instituto Nacional de Cancerologia. Bogotá, Colombia.*edevries@cancer.gov.co.

*(2) Dirección General Instituto Nacional de Cancerologia. Bogotá, Colombia.*

**References**

1. Bravo L, García L, Carrascal E, Rubiano J. Burden of breast cancer in Cali, Colombia: 1962-2012. Salud Publica Mex 2014;56(5):448-456. [ Links ]

2. Cortés A Bravo L, García L, Collazos R Incidencia, mortalidad y supervivencia por cáncer colorrectal en Cali, Colombia, 1962-2012. Salud Publica Mex 2014;56(5):457-464. [ Links ]

3. Restrepo J, Bravo L, García-Perdomo H, García L, Collazos P. Carbonell J. Incidencia, mortalidad y supervivencia al cáncer de próstata en Cali, Colombia, 1962-2011. Salud Publica Mex 2014; 56(5):440-447. [ Links ]

4. de Vries E, Karim-Kos HE Janssen-Heijnen ML, Soerjomataram I, Kiemeney LA, Coebergh JW. Explanations for worsening cancer survival. Nat Rev Clin Oncol 2010;7(1):60-63. [ Links ]

5. 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(3):326-335. [ Links ]

]]>6. Kleinbaum DG, Klein M. Survival Analysis: A Self-Learning Text, Third Edition. 3 ed. New York, Dordrecht, Heidelberg, London: Springer, 2012. [ Links ]

]]>