Timo Hakulinen: Assessing patients’ survival

A 1964 comparison (End Results of Cancer Therapy) of five-year survival rates for US and European patients showed that there was much scope for improvement in the Finnish figures. At that time, the processing of patient data in the registry was computerised and modern methods based on relative survival analysis were applied to the data. The publication was the first to compare population-based survival rates based on data for all patients in whole populations, including those who had not received treatment. The comparison focused on the performance of regional treatment systems in cancer care.

Relative survival analysis compares between populations the proportion of total patient mortality that exceeds the mortality of the general population that is comparable to patients. This excess mortality (over-mortality) is therefore considered to be related to patients’ cancer. Survival rates calculated on the basis of excess mortality are called relative survival rates. The general population in the area covered by the registry is set to be comparable to the patients in terms of sex, age and calendar period of follow-up.

For many cancers, the excess mortality rate for surviving patients decreases with longer follow-up from the time of cancer diagnosis. If the excess mortality rate disappears completely during the follow-up period, the surviving patients can be considered cured.

However, at follow-up periods longer than five years, unexpected results began to emerge: patients who survived appeared to have lower mortality than the general population. Some patients even reached very high ages and did not seem to die at all. Fortunately, the late 1960s saw the introduction of personal identity numbers and centralised population registration in Finland. It turned out that the earlier manual comparison had missed death information for some patients, and updating and correcting the data almost completely eliminated this puzzlement. However, for patients who had died before being assigned an ID, it was not possible to fully correct the problem.

But that was not the end of the problems. Two different methods of calculating relative survival rates were used, which gave widely divergent results as the follow-up period grew longer. Even if the patients did not have a lower mortality rate than the comparable general population, the 10-year relative survival rate might look higher than the 5-year rate. Methodological research was therefore needed.

As a result of this research, it was possible to calculate acceptable relative survival rates and compare them statistically. The methods were compiled into a statistical software package, the use of which was taught in courses at home, abroad and internationally. In total, hundreds of people were trained and the methods were widely used nationally and internationally. The development of statistical computing software packages made it possible to simultaneously include background factors in relative survival analyses by building regression models for excess mortality. The need for a separate software package was gradually eliminated, but the courses continued.

Over the years, there have been several major publications in Finland on the relative survival rates of patients, which have increased significantly over time. The figures also started to be published regularly in the annual statistics of the registry. In addition to the Nordic comparisons, there have been major international comparative studies in Europe and worldwide (EUROCARE, EUNICE, CONCORD), which have shown that the relative survival rates of Finnish patients are of a high international standard.

The comparison and interpretation of relative survival rates has been the subject of a strong and active research effort, resulting in, among other things, numerous doctoral theses and major articles. The methods found their way into textbooks and reference works worldwide. Research was often carried out in close international cooperation.

The methods have continued to evolve, and the long history and reliability of the Cancer Registry was able to demonstrate the practical feasibility of the new methods. A major innovation was the introduction of the so-called periodic method. This made it possible to calculate relative survival rates as far in advance as possible. When statistically modelled, these figures could be used in combination with cancer incidence and population projections to predict future cancer burden and mortality in the population. The periodic method was widely used internationally and future projections were made in a comparable way for all Nordic countries, among others.

Although survival rates for Finnish patients were high by international comparison, for many cancers there were differences in relative survival rates between regions and social groups in Finland, even when the general population was set to be similar in terms of region and social group. But the differences had decreased over time. Within-country differences were much larger in many other countries, such as Italy and the UK. Systematic differences allowed an estimate of the life-year savings that could be achieved if relative life expectancy were at the optimal level everywhere. It was possible to estimate the increase in uncertainty in relative life expectancy caused by the reduction in the size of comparable general population groups.

Statistical methods were developed for the survival of patients with multiple cancers to estimate the proportion of excess mortality from cancer for each patient. Of international importance was the study of the impact of incomplete cancer registration on relative survival rates and the possibility of correcting these rates. Methods were also developed to estimate the proportion of cured patients even before the whole patient population had been followed for so long that the survival rate had to consist only of cured patients.

There were also a number of issues that needed to be addressed or thought through on a case-by-case basis, and it was not easy to provide general guidance. For example, how to deal with cancers for which the only information was based on the death certificate? This was particularly important in international comparisons, where the proportion of such cases varied widely between regions and countries.

In many countries, data protection was and still is a problem that hampers research and reduces its quality and comparability. Some data were only available on condition that they were inaccurate or incomplete. For example, dates were not necessarily available in full. With complete Cancer Registry data, it was possible to assess the impact of any grossing-up or deletions made on the emergence of new systematic errors and to develop correction methods for erroneous figures.

The most successful way of obtaining clinical tumour and treatment data for the Cancer Registry was through individual studies, including international benchmarking projects and many doctoral theses. These were able to investigate and explain the differences observed, provided that many of the technical issues discussed above were properly taken into account.