Identifying causal relationships of cancer treatment and long-term health effects among 5-year survivors of childhood cancer in Southern Sweden

Survival after childhood cancer has improved but adverse so-called late effects are evident. Research on late effects can improve treatments and follow-up. Our study developed and used a method, based on AI techniques, to make causal associations between childhood cancer, treatments, and outcomes.
Published in Cancer
Like

Historical background Survival after childhood cancer was unusual before the 1950-ies. Certain diagnoses, like Wilm’s disease and Hodgkin’s lymphoma, had survival rates of 30-50% during the 1960-ies due to surgical interventions and irradiation therapy, but survival after childhood leukemia was largely unheard of before the 1970ies. A combination of chemotherapeutics, accompanied by prophylactic radiotherapy to the brain made a leukemia survival of 50% possible in the early 1980ies. Side effects of the curative treatments, so called late effects, emerged with time and became obvious to the clinical professions as survival increased.  The intense developments, including technical and care-related, which made an 80% survival possible across most childhood cancer diagnoses in the late 1990ies, continued to have disadvantages in the form of late effects.

Examples of historic (1-2) and contemporary (3-4) late effects

1) partial irradiation of the spine led to incomplete growth of affected areas resulting in a skewed posture.

2) radiotherapy to the brain to cure brain tumors or prophylactic radiotherapy of the brain in case of leukemia impacted cognition and endocrine functions.

3) breast cancer can occur in those who were treated with radiotherapy to the chest as part of their treatment at a young age. Second cancers are a well-known late effect.

4) chemotherapeutic agents and radiotherapy may affect fertility and cardiac function.

Current state The number of survivors and the follow-up time have increased resulting in increased attention of late effects. Today we know that 70% of all survivors will experience late effects and half of these will be of a serious or life-threatening nature 1. Premature mortality and high morbidity are evident although this is changing for patients treated more recently 2.  Three lines of development have resulted: 1) more research to discern what is enough and appropriate treatment to cure, yet limiting the late effects, 2) an increased awareness of late effects to the patients and their families. 3) an increased focus on altering primary treatment protocols to limit, or if possible avoid, late effects.

Data underlying research The largest childhood cancer diagnoses groups e.g. leukemia have received a great deal of focus as they are not only a major group but also easy to identify in registries by codes, and analyze statistically. Studies of more rare diagnoses has built on pooling of data and cross-border collaborations 3. Data acquisition has in many cases been a time consuming task due to obstacles with patients lost to follow-up and hard-to-find or fractionated historical medical charts. Despite these obstacles, large survivor cohorts have been assembled in regions where survival has reached high levels. Research in the Nordic countries has drawn on advantages of long standing national registries for outcomes 4, 5. Prominent research centers like St Jude Children’s Research Hospital draw on their historical lead on collection of structured data and directed childhood cancer survivorship research 6, 7. Pan-European efforts have gathered an unprecedented large number of survivors’ data for research 8, 9. All efforts have progressed the field and have proven highly relevant in laying new ground in improving health of childhood cancer survivors in terms of e.g. formulation and promotion of evidence based guidelines. Collectively, the work has also created a strong backdrop for our study.

Large-scale analysis of health care data is generating new findings with regard to patient-physician communication, side effect recognition, altered patient behavior, fewer re-admissions, and reduced costs to patients and society. Health care data has also sparked an interest from industry for potential improvements in clinical trial design, and drug safety surveillance. Big data and big data analytics however come with risks of biases, and skews, which may propagate into less useful results and create confusion. With limited information and rare occurrences, an interesting challenge presents itself: the ability of analyzing what actually is available. Based on the PC algorithm 10, 11, (named after the creators Peter Spirtes and Clark Glymour), we built a causal graph between all individual properties, childhood cancer diagnoses, cancer treatments, and potential late effects. To use the method with a limited amount of data, we developed a specially tailored conditional independence test, based on Bayesian estimation of distributions and a measure of correlation, which is invariant under conditioning. This makes it possible to base the estimation of the correlation on the whole data set also when conditioning on one or more variables, and thus makes it possible to maintain a high level of significance despite a limited amount of data.

The current work In the current article we examined causal associations between a history of childhood cancer and outcomes, by applying the developed method to a population based cohort of childhood cancer survivors12. We linked primary cancer diagnoses, predispositions and treatments i.e. chemotherapeutics, irradiation therapy doses, surgery and stem cell transplantation to outcomes from national registries. The resulting data set was small, rich in detail and was analyzed to identify causal associations. Cross-disciplinary discussions were a major part of the validation of the findings. We found five associations which to our knowledge have not been described before. These relate to eye conditions, fertility, and viral infections. Other identified associations (n=93) represent and confirm already well-established associations between a history of childhood cancer and late effects. Limitations and points of caution are discussed. The study also examined rates of health care usage and mortality - both results confirm what was previously known, higher rates were found for childhood cancer survivors as compared to the general population.

In conclusion, the study developed an AI-based approach to identify causal relationships in small data sets with potential for other areas, and led to new findings on the consequences of cancer treatment.  

References

  1. Oeffinger, K. C. et al. Chronic health conditions in adult survivors of childhood cancer. N. Engl. J. Med. 355, 1572-1582 (2006).
  2. Yeh, J. et al. Life Expectancy of Adult Survivors of Childhood Cancer Over 3 Decades. JAMA Oncology 6, 350-357 (2020).
  3. Gatta G, Capocaccia R, Botta L, Mallone S, De Angelis R, Ardanaz E, Comber H, Dimitrova N, Leinonen MK, Siesling S, van der Zwan JM, Van Eycken L, Visser O, Žakelj MP, Anderson, L.A. et al. Burden and centralised treatment in Europe of rare tumours: results of RARECAREnet-a population-based study.  Lancet Oncol. 18, 1022 (2017).
  4. Asdahl, P. H. et al. The Adult Life After Childhood Cancer in Scandinavia (ALiCCS) Study: Design and Characteristics. Pediatr. Blood Cancer. 62, 2204-2210 (2015).
  5. Winther, J. F. et al. Childhood cancer survivor cohorts in Europe. Acta Oncol. 54, 655-668 (2015).
  6. Robison, L. L. et al. Study design and cohort characteristics of the Childhood Cancer Survivor Study: a multi-institutional collaborative project. Med Pediatr Oncol. 38, 229 (2002).
  7. Howell, C. R. et al. Cohort Profile: The St. Jude Lifetime Cohort Study (SJLIFE) for paediatric cancer survivors. Int. J. Epidemiol. 50, 39-49 (2021).
  8. Byrne, J. et al. The PanCareSurFup consortium: research and guidelines to improve lives for survivors of childhood cancer. Eur. J. Cancer 103, 238-248 (2018).
  9. Grabow, D. et al. The PanCareSurFup cohort of 83,333 five-year survivors of childhood cancer: a cohort from 12 European countries. Eur. J. Epidemiol. (2018).
  10. Pearl, J. in Causality: Models, Reasoning, and Inference (Cambridge University Press, 2000).
  11. Spirtes, P., Glymour, C. & Scheines, R. in Causation, Prediction, and Search (MIT Press, Cambridge, MA, 2001).
  12. Wiebe, T., Hjorth, L., Marotta Kelly, M., Linge, H. M. & Garwicz, S. A population based pediatric oncology registry in Southern Sweden: the BORISS registry. Eur. J. Epidemiol. 33, 1125-1129 (2018).

 

Please sign in or register for FREE

If you are a registered user on Research Communities by Springer Nature, please sign in

Subscribe to the Topic

Cancer Biology
Life Sciences > Biological Sciences > Cancer Biology

Related Collections

With collections, you can get published faster and increase your visibility.

Precision medicine in diabetes

This Collection includes a series of systematic reviews published by the ADA/EASD Precision Medicine in Diabetes Initiative consortium and is also open to other submissions on precision diabetes medicine.

Publishing Model: Open Access

Deadline: Apr 25, 2024

Liquid biopsy

This Collection welcomes clinical and translational research on liquid biopsy approaches in cancer.

Publishing Model: Open Access

Deadline: May 13, 2024