Loss in life expectancy and gain in life years as measures of cancer impact

An empirical approach for life expectancy estimation based on survival analysis among a post-acute myocardial infarction population

Background

Practical communication of prognosis is pertinent in the clinical setting. Survival analysis techniques are standardly used in cohort studies; however, their results are not straightforward for interpretation as compared to the graspable notion of life expectancy (LE). The present study empirically examines the relationship between Cox regression coefficients (HRs), which reflect the relative risk of the investigated risk factors for mortality, and years of potential life lost (YPLL) values after acute myocardial infarction (AMI).

Methods

This retrospective population-based study included patients aged 40-80 years, who survived AMI hospitalization from January 1, 2002, to October 25, 2017. A survival analysis approach assessed relationships between variables and the risk for all-cause mortality in an up to 21-year follow-up period. The total score was calculated for each patient as the summation of the Cox regression coefficients (AdjHRs) values. Individual LE and YPLL were calculated. YPLL was assessed as a function of the total score.

Results

The cohort (n = 6316, age 63.0 ± 10.5 years, 73.4 % males) was randomly split into training (n = 4243) and validation (n = 2073) datasets. Sixteen main clinical risk factors for mortality were explored (total score of 0-14.2 points). After adjustment for age, sex and nationality, a one-point increase in the total score was associated with YPLL of ∼one year. A goodness-of-fit of the prediction model found 0.624 and 0.585 for the training and validation datasets respectively.

Conclusions

This functional derivation for converting coefficients of survival analysis into the comprehensible form of YPLL/LE allows for practical prognostic calculation and communication.

Expected life years compared to the general population

For cohorts with long-term follow-up, the number of years lost due to a certain disease yields a measure with a simple and appealing interpretation. Recently, an overview of the methodology used for this goal has been published, and two measures have been proposed. In this work, we consider a third option that may be useful in settings in which the other two are inappropriate. In all three measures, the survival of the given dataset is compared to the expected survival in the general population which is calculated using external mortality tables. We thoroughly analyze the differences between the three measures, their assumptions, interpretation, and the corresponding estimators. The first measure is defined in a competing risk setting and assumes an excess hazard compared to the population, while the other two measures also allow estimation for groups that live better than the general population. In this case, the observed survival of the patients is compared to that in the population. The starting point of this comparison depends on whether the entry into the study is a hazard changing event (e.g., disease diagnosis or the age at which the inclusion criteria were met). Focusing on the newly defined life years difference measure, we study the estimation of the variance and consider the possible challenges (e.g., extrapolation) that occur in practice. We illustrate its use with a dataset of French Olympic athletes. Finally, an efficient R implementation has been developed for all three measures which make this work easily available to subsequent users.

Generating high-fidelity synthetic time-to-event datasets to improve data transparency and accessibility

BACKGROUND

A lack of available data and statistical code being published alongside journal articles provides a significant barrier to open scientific discourse, and reproducibility of research. Information governance restrictions inhibit the active dissemination of individual level data to accompany published manuscripts. Realistic, high-fidelity time-to-event synthetic data can aid in the acceleration of methodological developments in survival analysis and beyond by enabling researchers to access and test published methods using data similar to that which they were developed on.

METHODS

We present methods to accurately emulate the covariate patterns and survival times found in real-world datasets using synthetic data techniques, without compromising patient privacy. We model the joint covariate distribution of the original data using covariate specific sequential conditional regression models, then fit a complex flexible parametric survival model from which to generate survival times conditional on individual covariate patterns. We recreate the administrative censoring mechanism using the last observed follow-up date information from the initial dataset. Metrics for evaluating the accuracy of the synthetic data, and the non-identifiability of individuals from the original dataset, are presented.

RESULTS

We successfully create a synthetic version of an example colon cancer dataset consisting of 9064 patients which aims to show good similarity to both covariate distributions and survival times from the original data, without containing any exact information from the original data, therefore allowing them to be published openly alongside research.

CONCLUSIONS

We evaluate the effectiveness of the methods for constructing synthetic data, as well as providing evidence that there is minimal risk that a given patient from the original data could be identified from their individual unique patient information. Synthetic datasets using this methodology could be made available alongside published research without breaching data privacy protocols, and allow for data and code to be made available alongside methodological or applied manuscripts to greatly improve the transparency and accessibility of medical research.

Reference adjusted loss in life expectancy for population-based cancer patient survival comparisons - with an application to colon cancer in Sweden

BACKGROUND

The loss in life expectancy, LLE, is defined as the difference in life expectancy between cancer patients and that of the general population. It is a useful measure for summarising the impact of a cancer diagnosis on an individual's life expectancy. However, it is less useful for making comparisons of cancer survival across groups or over time, since the LLE is influenced by both mortality due to cancer and other causes and the life expectancy in the general population.

METHODS

We present an approach for making LLE estimates comparable across groups and over time by using reference expected mortality rates with flexible parametric relative survival models. The approach is illustrated by estimating temporal trends in LLE of colon cancer patients in Sweden.

RESULTS

The life expectancy of Swedish colon cancer patients has improved, but the LLE has not decreased to the same extent since the life expectancy in the general population has also increased. When using a fixed population and other-cause mortality, i.e. a reference-adjusted approach, the LLE decreases over time. For example, using 2010 mortality rates as the reference, the LLE for females diagnosed at age 65 decreased from 11.3 if diagnosed in 1976 to 7.2 if diagnosed in 2010, and from 3.9 to 1.9 years for women 85 years old at diagnosis.

CONCLUSION

The reference-adjusted LLE is useful for making comparisons across calendar time, or groups, since differences in other cause mortality are removed.

IMPACT

The reference-adjusted approach enhances the use of LLE as a comparative measure.

Quantifying Differences in Remaining Life Expectancy after Cancer Diagnosis, Aboriginal and Torres Strait Islanders, and Other Australians, 2005-2016

BACKGROUND

This study quantified differences in remaining life expectancy (RLE) among Aboriginal and Torres Strait Islander and other Australian patients with cancer. We assessed how much of this disparity was due to differences in cancer and noncancer mortality and calculated the population gain in life years for Aboriginal and Torres Strait Islanders cancer diagnoses if the cancer survival disparities were removed.

METHODS

Flexible parametric relative survival models were used to estimate RLE by Aboriginal and Torres Strait Islander status for a population-based cohort of 709,239 persons (12,830 Aboriginal and Torres Strait Islanders), 2005 to 2016.

RESULTS

For all cancers combined, the average disparity in RLE was 8.0 years between Aboriginal and Torres Strait Islanders (12.0 years) and other Australians (20.0 years). The magnitude of this disparity varied by cancer type, being >10 years for cervical cancer versus <2 years for lung and pancreatic cancers. For all cancers combined, around 26% of this disparity was due to differences in cancer mortality and 74% due to noncancer mortality. Among 1,342 Aboriginal and Torres Strait Islanders diagnosed with cancer in 2015 an estimated 2,818 life years would be gained if cancer survival disparities were removed.

CONCLUSIONS

A cancer diagnosis exacerbates the existing disparities in RLE among Aboriginal and Torres Strait Islanders. Addressing them will require consideration of both cancer-related factors and those contributing to noncancer mortality.

IMPACT

Reported survival-based measures provided additional insights into the overall impact of cancer over a lifetime horizon among Aboriginal and Torres Strait Islander peoples.

Life expectancy and cancer survival in Oncosalud: outcomes over a 15-year period in a Peruvian private institution

Background

There is a large gap in the data on cancer outcomes in Latin America, making it difficult to establish adequate cancer control policies in the region. The aim of our study was to describe the survival, life expectancy estimates and life expectancy changes over time for a large cohort of Peruvian patients insured with Oncosalud, a private healthcare system.

Patients and methods

We evaluated a retrospective cohort of patients diagnosed between 2000 and 2015 in Oncosalud (Lima-Peru). Cases included colon, rectum, stomach, bladder, breast, prostate and non-melanoma skin cancers. Survival was evaluated with the Kaplan–Meier methodology. The standard period life table was used to estimate the excess mortality risks of patients in our cohort compared to the population covered by the Peruvian Superintendence of Banks, Insurance Companies and Pension Funds (SBS). The years of life lost was estimated based on SBS population, matching patients by age and sex.

Results

A large cohort of 7,687 Peruvian cancer patients managed in a 15-year period was eligible. If patients survive 5 years after a cancer diagnosis, life expectancy tends to be close to that of a population without cancer. The number of years of life lost at diagnosis was higher at the youngest ages, steadily decreasing thereafter. During the first years after cancer diagnosis, young patients face a much higher loss in life expectancy than older ones. Patients suffering from colon, rectum, stomach and bladder cancer are the most affected by the years of life lost.

Conclusion

In cancer patients surviving ≥ 5 years, life expectancy becomes similar to that observed in a population with similar socioeconomic characteristics. The estimated survival rate in our cohort is higher than that reported by public cancer registries in Peru. This could be explained by the different socio-economic background and access to specialised cancer care.

Dynamic prediction and analysis based on restricted mean survival time in survival analysis with nonproportional hazards

BACKGROUND AND OBJECTIVE

In the process of clinical diagnosis and treatment, the restricted mean survival time (RMST), which reflects the life expectancy of patients up to a specified time, can be used as an appropriate outcome measure. However, the RMST only calculates the mean survival time of patients within a period of time after the start of follow-up and may not accurately portray the change in a patient's life expectancy over time.

METHODS

The life expectancy can be adjusted for the time the patient has already survived and defined as the conditional restricted mean survival time (cRMST). A dynamic RMST model based on the cRMST can be established by incorporating time-dependent covariates and covariates with time-varying effects. We analyzed data from a study of primary biliary cirrhosis (PBC) to illustrate the use of the dynamic RMST model, and a simulation study was designed to test the advantages of the proposed approach. The predictive performance was evaluated using the C-index and the prediction error.

RESULTS

Considering both the example results and the simulation results, the proposed dynamic RMST model, which can explore the dynamic effects of prognostic factors on survival time, has better predictive performance than the RMST model. Three PBC patient examples were used to illustrate how the predicted cRMST changed at different prediction times during follow-up.

CONCLUSIONS

The use of the dynamic RMST model based on the cRMST allows for the optimization of evidence-based decision-making by updating personalized dynamic life expectancy for patients.

Temporal improvements noted in life expectancy of patients with colorectal cancer; a Dutch population-based study

OBJECTIVE

Specific survival estimates are needed for the increasing number of colorectal cancer (CRC) survivors. The aim of this population-based study was to determine conditional loss in expectation of life (LEL) due to CRC.

STUDY DESIGN AND SETTING

All surgically treated patients with CRC registered in the Netherlands Cancer Registry with stage I-III between 1990-2016, were included (N = 203,216). Estimates of conditional LEL were predicted using flexible parametric models and the total life years lost due to cancer were estimated.

RESULTS

LEL decreased with older age and patients with rectal cancer or higher disease stage had highest LEL. In 2010, LEL for sixty-year old male and female patients was 2 vs. 2, 4 vs. 4, and 7 vs. 8 years for colon cancer, and 2 vs. 2, 4 vs. 5 and 7 vs. 8 years for rectal cancer, respectively. Conditional LEL in patients with CRC decreased during follow-up. Patients with combined stage I-III colon and rectal cancer in 2010 lost an estimated 18,628 and 11,336 life years.

CONCLUSION

This study quantified the impact of CRC on patient's life expectancy, both on individual and population level and demonstrated temporal improvements in CRC survival. These results provide meaningful information that can be used during follow-up.

Cancer survival statistics for patients and healthcare professionals - a tutorial of real-world data analysis

Monitoring survival of cancer patients using data collected by population-based cancer registries is an important component of cancer control. In this setting, patient survival is often summarized using net survival, that is survival from cancer if there were no other possible causes of death. Although net survival is the gold standard for comparing survival between groups or over time, it is less relevant for understanding the anticipated real-world prognosis of patients. In this review, we explain statistical concepts targeted towards patients, clinicians and healthcare professionals that summarize cancer patient survival under the assumption that other causes of death exist. Specifically, we explain the appropriate use, interpretation and assumptions behind statistical methods for competing risks, loss in life expectancy due to cancer and conditional survival. These concepts are relevant when producing statistics for risk communication between physicians and patients, planning for use of healthcare resources, or other applications when consideration of both cancer outcomes and the competing risks of death is required. To reinforce the concepts, we use Swedish population-based data of patients diagnosed with cancer of the breast, prostate, colon and chronic myeloid leukaemia. We conclude that when choosing between summary measures of survival it is critical to characterize the purpose of the study and to determine the nature of the hypothesis under investigation. The choice of terminology and style of reporting should be carefully adapted to the target audience and may range from summaries for specialist readers of scientific publications to interactive online tools aimed towards lay persons.

Impact of area-level socioeconomic status and accessibility to treatment on life expectancy after a cancer diagnosis in Queensland, Australia

AIMS

This study quantifies geographic inequities in loss of life expectancy (LOLE) by area-level socioeconomic status (SES) and accessibility to treatment.

METHODS

Analysis was conducted using a population-based cancer-registry cohort (n = 371,570) of Queensland (Australia) residents aged 50-89 years, diagnosed between 1997-2016. Flexible parametric survival models were used to estimate LOLE by area-level SES and accessibility for all invasive cancers and the five leading cancers. The gain in life years that could be achieved if all cancer patients experienced the same relative survival as those in the least disadvantaged-high accessibility category was estimated for the 2016 cohort.

RESULTS

For all invasive cancers, men living in the most disadvantaged areas lost 34 % of life expectancy due to their cancer diagnosis, while those from the least disadvantaged areas lost 25 %. The corresponding percentages for women were 33 % and 23 %. Accessibility had a lower impact on LOLE than SES, with patients from low accessibility areas losing 0-4 % more life expectancy than those from high accessibility areas. For cancer patients diagnosed in 2016 (n = 24,423), an estimated 101,387 life years will be lost. This would be reduced by 19 % if all patients experienced the same relative survival as those from the least disadvantaged-high accessibility areas.

CONCLUSION

The impact of a cancer diagnosis on remaining life expectancy varies by geographical area. Establishing reasons why area disadvantage impacts on life expectancy is crucial to inform subsequent interventions that could increase the life expectancy of cancer patients from more disadvantaged areas.

Quantifying the Number of Cancer Deaths Avoided Due to Improvements in Cancer Survival since the 1980s in the Australian Population, 1985-2014

BACKGROUND

This study quantifies the number of potentially "avoided"cancer deaths due to differences in 10-year relative survival between three time periods, reflecting temporal improvements in cancer diagnostic and/or treatment practices in Australia.

METHODS

National population-based cohort of 2,307,565 Australians ages 15 to 89 years, diagnosed with a primary invasive cancer from 1985 to 2014 with mortality follow-up to December 31, 2015. Excess mortality rates and crude probabilities of cancer deaths were estimated using flexible parametric relative survival models. Crude probabilities were then used to calculate "avoided cancer deaths" (reduced number of cancer deaths within 10 years of diagnosis due to survival changes since 1985-1994) for all cancers and 13 leading cancer types.

RESULTS

For each cancer type, excess mortality (in the cancer cohort vs. the expected population mortality) was significantly lower for more recently diagnosed persons. For all cancers combined, the number of "avoided cancer deaths" (vs. 1985-1994) was 4,877 (1995-2004) and 11,385 (2005-2014) among males. Prostate (1995-2004: 2,144; 2005-2014: 5,099) and female breast cancer (1,127 and 2,048) had the highest number of such deaths, whereas <400 were avoided for pancreatic or lung cancers across each period.

CONCLUSIONS

Screening and early detection likely contributed to the high number of "avoided cancer deaths" for prostate and female breast cancer, whereas early detection remains difficult for lung and pancreatic cancers, highlighting the need for improved preventive and screening measures.

IMPACT

Absolute measures such as "avoided cancer deaths" can provide a more tangible estimate of the improvements in cancer survival than standard net survival measures.

Does minimum follow-up time post-diagnosis matter? An assessment of changing loss of life expectancy for people with cancer in Western Australia from 1982 to 2016

BACKGROUND

Cancer survival has improved in Western Australia (WA) over recent decades. Loss of life expectancy (LOLE) is a useful measure for assessing cancer survival at a population-level. Some previous studies estimating LOLE have required a minimum follow-up beyond diagnosis to reduce the impact of modelled extrapolation, while others have not. The first aim of this study was to assess the impact of minimum length of follow-up on LOLE estimates for people diagnosed in 2006 with female breast, colorectal, prostate, lung, cervical, combined oesophageal and stomach cancers, and melanoma. Based on these results, the second aim was to assess temporal changes in LOLE for these cancer types for diagnoses between 1982 and 2016.

METHODS

Person-level linked cancer registry and mortality data were used for invasive primary cancer diagnoses for WA residents aged 15-89 years. The analysis for aim one included cases diagnosed from 1982 to the end of 2006, followed to the end of 2006 (i.e. no minimum follow-up), 2011 (i.e. five years minimum follow-up, assuming survival) or 2016 (i.e. 10 years minimum follow-up). To achieve the second study aim, the diagnostic period was extended to the end of 2016. Life expectancy estimates were obtained after fitting flexible parametric relative survival models. Single-year age and sex-specific death rates were used as a reference to estimate LOLE and proportionate loss of life expectancy.

RESULTS

Temporal changes were not reported for prostate, cervical, oesophageal and stomach cancers or melanoma, due to differences in LOLE estimates by minimum follow-up time, or estimate imprecision. Marked reductions in LOLE were observed for female breast and colorectal cancer. There was minimal absolute reduction for lung cancer, where LOLE remained high.

CONCLUSION

This study considered the appropriateness of including recent cancer diagnoses when assessing temporal changes in LOLE, finding variation in estimates with differing minimum follow-up or high parameter uncertainty for most included cancer types. Temporal changes in LOLE in-turn reflected changes in the life expectancy of the general population, cancer detection and management. These factors must be considered when estimating and interpreting LOLE estimates.

Illustration of different modelling assumptions for estimation of loss in expectation of life due to cancer

BACKGROUND

The life expectancy of cancer patients, and the loss in expectation of life as compared to the life expectancy without cancer, is a useful measure of cancer patient survival and complement the more commonly reported 5-year survival. The estimation of life expectancy and loss in expectation of life generally requires extrapolation of the survival function, since the follow-up is not long enough for the survival function to reach 0. We have previously shown that the survival of the cancer patients can be extrapolated by breaking down the all-cause survival into two component parts, the expected survival and the relative survival, and make assumptions for extrapolation of these functions independently. When extrapolating survival from a model including covariates such as calendar year, age at diagnosis and deprivation status, care has to be taken regarding the assumptions underlying the extrapolation. There are often different alternative ways for modelling covariate effects or for assumptions regarding the extrapolation.

METHODS

In this paper we describe and discuss different alternative approaches for extrapolation of survival when estimating life expectancy and loss in expectation of life for cancer patients. Flexible parametric models within a relative survival setting are used, and examples are presented using data on colon cancer in England.

RESULTS

Generally, the different modelling assumptions and approaches give small differences in the estimates of loss in expectation of life, however, the results can differ for younger ages and for conditional estimates.

CONCLUSION

Sensitivity analyses should be performed to evaluate the effect of the assumptions made when modelling and extrapolating survival to estimate the loss in expectation of life.