pith. sign in

arxiv: 2605.31213 · v2 · pith:BHVZWCYHnew · submitted 2026-05-29 · 💻 cs.NE

Developing a novel Comorbidities Index for predicting 10-year mortality in Prostate Cancer patients: A computational data-driven approach

Davide Farinati , Francesco Barletta , Paolo Zaurito , Simone Scuderi , Nicholas Raison , Alejandro Granados , Prokar Dasgupta , Giorgio Gandaglia
show 1 more author
Alberto Briganti
This is my paper

Pith reviewed 2026-06-28 20:05 UTC · model grok-4.3

classification 💻 cs.NE
keywords comorbidities indexprostate cancerten-year mortalitydata-driven approachbio-inspired algorithmsradical prostatectomysurvival predictionconcordance index
0
0 comments X

The pith

Data-driven optimization creates a comorbidities index that predicts ten-year mortality more accurately in prostate cancer patients eligible for surgery.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper establishes a framework for deriving a prostate cancer-specific comorbidities index by applying bio-inspired algorithms to recalibrate weights and generate new symbolic models from patient data. This matters for clinical practice because radical prostatectomy is only recommended when patients have at least ten years of life expectancy, so better estimates of other-cause mortality help avoid unnecessary treatments. The optimized approaches including genetic algorithms and particle swarm variants show improved performance over the standard Charlson index, particularly when incorporating cancer-specific factors. The resulting tool offers a more accurate and interpretable way to assess competing risks in treatment decisions.

Core claim

Using a retrospective single-institution cohort of prostate cancer patients considered for radical prostatectomy, population-based bio-inspired algorithms recalibrate the weights of the Charlson Comorbidities Index and evolve alternative formulations to better discriminate ten-year survival, with methods such as GA, FST-PSO, and SLIM achieving higher concordance indices than the original CCI or the prostate cancer-specific PCCI, especially when prostate cancer variables are included.

What carries the argument

Population-Based Bio-Inspired Algorithms (PBBIAs) applied to recalibrate comorbidity weights and evolve symbolic formulations for optimized ten-year survival discrimination.

If this is right

  • GA, FST-PSO, and SLIM outperform both the original CCI and the PCCI when PCa-specific variables are included.
  • These methods improve the Concordance Index by up to 0.1.
  • GPLearn produces compact and interpretable models with competitive performance.
  • The approach yields an updated and interpretable tool for improving patient selection for radical prostatectomy.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the new index generalizes, it could reduce overtreatment by better identifying patients unlikely to live ten years due to comorbidities.
  • Similar optimization techniques might be applied to create tailored indices for other cancer types where competing mortality risks affect treatment choices.
  • Integration of additional clinical variables beyond comorbidities could further refine the predictions in future extensions.

Load-bearing premise

The retrospective single-institution cohort used for model development and testing represents the wider population of prostate cancer patients considered for radical prostatectomy.

What would settle it

A test on an independent external dataset from multiple institutions where the new indices do not show a statistically significant improvement in concordance index over the original CCI would indicate the claimed superiority does not hold.

Figures

Figures reproduced from arXiv: 2605.31213 by Alberto Briganti, Alejandro Granados, Davide Farinati, Francesco Barletta, Giorgio Gandaglia, Nicholas Raison, Paolo Zaurito, Prokar Dasgupta, Simone Scuderi.

Figure 1
Figure 1. Figure 1: Test C-index comparison across all methods on the standard dataset. [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Two dimensional plot displaying the trade-off between performance (Test C [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Test C-index comparison across all methods when including PCa specific vari [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Model size comparison (number of nodes) on the urological variables dataset. [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Two dimensional plot displaying the trade-off between performance (Test C [PITH_FULL_IMAGE:figures/full_fig_p013_5.png] view at source ↗
read the original abstract

The Charlson Comorbidities Index (CCI) is a weighted additive index widely used to estimate ten-year mortality risk, but its original weights may not reflect contemporary prognoses. This limitation is critical in Prostate Cancer (PCa), where radical treatment is recommended only for patients with a life expectancy of at least ten years. For candidates eligible for Radical Prostatectomy (RP), accurate estimation of ten-year other-cause mortality is essential to balance oncological benefit against competing risks and avoid overtreatment. We propose a data-driven framework to derive a comorbidity index tailored to PCa patients considered for RP. Using a retrospective single-institution cohort, we apply Population-Based Bio-Inspired Algorithms (PBBIAs) to recalibrate comorbidity weights and evolve alternative symbolic formulations optimized for ten-year survival discrimination. We compared six optimization strategies, including symbolic regression approaches based on Genetic Programming (GP), population-based metaheuristics, clinically validated baselines, and survival prediction models. Results show that GA, FST-PSO, and SLIM outperform both the original CCI and the PCCI, particularly when PCa-specific variables are included, improving the Concordance Index by up to 0.1. GPLearn yields compact and interpretable models with competitive performance. Overall, the proposed approach provides an updated and interpretable tool to improve patient selection for RP.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 1 minor

Summary. The paper claims that Population-Based Bio-Inspired Algorithms (PBBIAs) including GA, FST-PSO, and SLIM can be used on a single retrospective single-institution cohort of prostate cancer patients considered for radical prostatectomy to recalibrate comorbidity weights and evolve symbolic forms, yielding indices that outperform the original CCI and PCCI (with C-index gains up to 0.1) especially when PCa-specific variables are added; GPLearn is highlighted for producing compact interpretable models.

Significance. If the performance claims survive proper out-of-sample validation, the work would supply a computationally derived, potentially more accurate tool for estimating 10-year other-cause mortality to guide RP eligibility and reduce overtreatment. The application of evolutionary algorithms both for weight optimization and for symbolic regression in a survival setting is a methodological contribution worth noting.

major comments (3)
  1. [Abstract] Abstract: the reported C-index improvements (up to 0.1) are obtained by applying GA, FST-PSO, and SLIM to recalibrate weights and evolve expressions on the identical retrospective cohort used to compute the concordance indices; no cross-validation, temporal split, or external validation set is described, so the observed lift cannot be separated from in-sample fitting.
  2. [Abstract] Abstract and methods description: the central claim that the new indices 'outperform' CCI and PCCI rests on discrimination measured after optimization on the same ten-year survival outcomes; this circularity means the gains are expected by construction and do not yet demonstrate improved generalization.
  3. [Abstract] Abstract: the single-institution retrospective cohort is used for both model derivation and performance reporting, yet no information is supplied on cohort size, event rate, follow-up completeness, or any control for overfitting (e.g., nested CV or regularization), leaving the robustness of the 0.1 C-index delta unsupported.
minor comments (1)
  1. [Abstract] Abstract: the phrase 'six optimization strategies' is used but only four are named; list all six explicitly for clarity.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the detailed and constructive comments on validation, circularity, and cohort description. We agree that the original abstract was insufficiently explicit on these points and have revised the abstract, methods, and discussion to incorporate the necessary clarifications and limitations. Point-by-point responses are provided below.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the reported C-index improvements (up to 0.1) are obtained by applying GA, FST-PSO, and SLIM to recalibrate weights and evolve expressions on the identical retrospective cohort used to compute the concordance indices; no cross-validation, temporal split, or external validation set is described, so the observed lift cannot be separated from in-sample fitting.

    Authors: We agree that the abstract did not describe any cross-validation, temporal split, or external validation. The revised abstract now explicitly states that the reported C-index values reflect performance on the derivation cohort. We have added a dedicated limitations paragraph acknowledging that the observed gains are in-sample and that external validation on independent data is required before clinical use. No new external cohort is available for the current revision. revision: yes

  2. Referee: [Abstract] Abstract and methods description: the central claim that the new indices 'outperform' CCI and PCCI rests on discrimination measured after optimization on the same ten-year survival outcomes; this circularity means the gains are expected by construction and do not yet demonstrate improved generalization.

    Authors: The referee correctly identifies that optimization on the same outcomes will produce improved discrimination by construction. The revised text now frames the contribution as the derivation of alternative weightings and symbolic expressions via PBBIAs rather than as a claim of superior generalization. We have added language clarifying that prospective or external validation is needed to establish transportability and have tempered the abstract claims accordingly. revision: yes

  3. Referee: [Abstract] Abstract: the single-institution retrospective cohort is used for both model derivation and performance reporting, yet no information is supplied on cohort size, event rate, follow-up completeness, or any control for overfitting (e.g., nested CV or regularization), leaving the robustness of the 0.1 C-index delta unsupported.

    Authors: We accept that the original abstract omitted these essential descriptors. The revised abstract now reports cohort size, number of events, median follow-up, and completeness. In the methods we have added explicit description of the regularization terms used within the PBBIA objective functions and the bootstrap resampling procedure employed to assess stability of the C-index estimates. revision: yes

standing simulated objections not resolved
  • External validation on an independent multi-institutional cohort is not available in the current study.

Circularity Check

1 steps flagged

Comorbidity weights and symbolic forms optimized on same cohort used to report C-index gains, making reported improvements by construction

specific steps
  1. fitted input called prediction [Abstract]
    "Using a retrospective single-institution cohort, we apply Population-Based Bio-Inspired Algorithms (PBBIAs) to recalibrate comorbidity weights and evolve alternative symbolic formulations optimized for ten-year survival discrimination. ... Results show that GA, FST-PSO, and SLIM outperform both the original CCI and the PCCI, particularly when PCa-specific variables are included, improving the Concordance Index by up to 0.1."

    The optimization target is ten-year survival discrimination (the same quantity later measured by C-index). Weights and symbolic forms are fitted to the cohort's outcomes; the reported C-index lift is therefore the direct output of that fitting procedure on the identical data, not an independent test of generalization.

full rationale

The paper applies PBBIAs (GA, FST-PSO, SLIM) to recalibrate weights and evolve symbolic expressions explicitly optimized for ten-year survival discrimination on a single retrospective cohort, then reports that these yield up to 0.1 C-index improvement over CCI/PCCI on that same cohort. No external validation, temporal split, or hold-out is described, so the performance delta reduces to the result of fitting rather than independent prediction. This matches the fitted_input_called_prediction pattern with load-bearing impact on the central claim.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central claim depends on fitting parameters directly to the cohort survival data and on the untested assumption that the single-center retrospective sample represents the target clinical population.

free parameters (2)
  • comorbidity weights
    Recalibrated by GA, FST-PSO and other algorithms on the study cohort survival data.
  • symbolic regression parameters
    Evolved by genetic programming to maximize concordance on the same dataset.
axioms (1)
  • domain assumption The single-institution retrospective cohort is representative of the broader population of men considered for radical prostatectomy.
    Invoked when claiming the optimized index will improve patient selection.

pith-pipeline@v0.9.1-grok · 5806 in / 1391 out tokens · 32009 ms · 2026-06-28T20:05:26.600436+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

19 extracted references · 17 canonical work pages

  1. [1]

    In: Proceed- ings of the 2020 Genetic and Evolutionary Computation Conference Com- panion

    Bogdan Burlacu, Gabriel Kronberger, and Michael Kommenda. Operon c++: An efficient genetic programming framework for symbolic regres- sion. InProceedings of the 2020 Genetic and Evolutionary Computation Conference Companion, GECCO ’20, page 1562–1570, New York, NY, USA, 2020. Association for Computing Machinery. ISBN 9781450371278. https://doi.org/10.1145...

    work page doi:10.1145/3377929.3398099 2020

  2. [2]

    Szatrowski, Janey Peterson, and Jeffrey Gold

    Mary Charlson, Ted P. Szatrowski, Janey Peterson, and Jeffrey Gold. Val- idation of a combined comorbidity index.Journal of Clinical Epidemiol- ogy, 47(11):1245–1251, Nov 1994.https://doi.org/https://doi.org/ 10.1016/0895-4356(94)90129-5. URLhttps://pubmed.ncbi.nlm.nih. gov/7722560/

    work page doi:10.1016/0895-4356(94)90129-5 1994

  3. [3]

    Charlson, Peter Pompei, Kathy L

    Mary E. Charlson, Peter Pompei, Kathy L. Ales, and C.Ronald MacKenzie. A new method of classifying prognostic comorbidity in longitudinal studies: Development and validation.Journal of Chronic Diseases, 40(5):373–383, Jan 1987.https://doi.org/https://doi.org/10.1016/0021-9681(87) 90171-8. URLhttps://pubmed.ncbi.nlm.nih.gov/3558716/

    work page doi:10.1016/0021-9681(87 1987

  4. [4]

    Philip Cornford, Roderick C.N. van den Bergh, Erik Briers, Thomas Van den Broeck, Oliver Brunckhorst, Julie Darraugh, Daniel Eberli, Gert De Meerleer, Maria De Santis, Andrea Farolfi, Giorgio Gandaglia, Silke Gillessen, Nikolaos Grivas, Ann M. Henry, Michael Lardas, Geert J.L.H. van Leenders, Matthew Liew, Estefania Linares Espinos, Jan Oldenburg, Inge M....

    work page doi:10.1016/j.eururo.2024.03.027 2024

  5. [5]

    Cox and D

    D.R. Cox and D. Oakes. Analysis of survival data. Feb 2018. https://doi.org/https://doi.org/10.1201/9781315137438. URLhttps://www.taylorfrancis.com/books/mono/10.1201/ 9781315137438/analysis-survival-data-reid-cox

    work page doi:10.1201/9781315137438 2018

  6. [6]

    Daskivich, Lorna Kwan, Atreya Dash, Christopher Saigal, and Mark S

    Timothy J. Daskivich, Lorna Kwan, Atreya Dash, Christopher Saigal, and Mark S. Litwin. An age adjusted comorbidity index to predict long- term, other cause mortality in men with prostate cancer.Journal of Urology, 194(1):73–78, Jul 2015.https://doi.org/https://doi.org/10. 1016/j.juro.2015.01.081. URLhttps://pubmed.ncbi.nlm.nih.gov/ 25623745/

    2015

  7. [7]

    A survey on dynamic popula- tions in bio-inspired algorithms.Genetic Programming and Evolvable Ma- 16 D

    Davide Farinati and Leonardo Vanneschi. A survey on dynamic popula- tions in bio-inspired algorithms.Genetic Programming and Evolvable Ma- 16 D. Farinati et al. chines, 25(2), Jul 2024.https://doi.org/https://doi.org/10.1007/ s10710-024-09492-4. URLhttps://link.springer.com/article/10. 1007/s10710-024-09492-4

    2024

  8. [8]

    A study on the dynamics and effectiveness of the deflate geometric semantic muta- tion.IEEE Transactions on Evolutionary Computation, page 1–1, 2025

    Davide Farinati, Gloria Pietropolli, and Leonardo Vanneschi. A study on the dynamics and effectiveness of the deflate geometric semantic muta- tion.IEEE Transactions on Evolutionary Computation, page 1–1, 2025. https://doi.org/https://doi.org/10.1109/tevc.2025.3611226. URL https://ieeexplore.ieee.org/abstract/document/11168418

    work page doi:10.1109/tevc.2025.3611226 2025

  9. [9]

    Hamdy, Jenny L

    Freddie C. Hamdy, Jenny L. Donovan, J. Athene Lane, Malcolm Mason, Chris Metcalfe, Peter Holding, Michael Davis, Tim J. Peters, Emma L. Turner, Richard M. Martin, Jon Oxley, Mary Robinson, John Staffurth, Eleanor Walsh, Prasad Bollina, James Catto, Andrew Doble, Alan Doherty, David Gillatt, and Roger Kockelbergh. 10-year outcomes after monitoring, surgery...

    work page doi:10.1056/nejmoa1606220 2016

  10. [10]

    Kogalur, Eugene H

    Hemant Ishwaran, Udaya B. Kogalur, Eugene H. Blackstone, and Michael S. Lauer. Random survival forests.The Annals of Applied Statistics, 2(3), Sep 2008.https://doi.org/https: //doi.org/10.1214/08-aoas169. URLhttps://projecteuclid. org/journals/annals-of-applied-statistics/volume-2/issue-3/ Random-survival-forests/10.1214/08-AOAS169.full

    work page doi:10.1214/08-aoas169 2008

  11. [11]

    Katzman, Uri Shaham, Alexander Cloninger, Jonathan Bates, Tingting Jiang, and Yuval Kluger

    Jared L. Katzman, Uri Shaham, Alexander Cloninger, Jonathan Bates, Tingting Jiang, and Yuval Kluger. Deepsurv: personalized treatment recommender system using a cox proportional hazards deep neural net- work.BMC Medical Research Methodology, 18(1), Feb 2018.https:// doi.org/https://doi.org/10.1186/s12874-018-0482-1. URLhttps: //link.springer.com/article/1...

    work page doi:10.1186/s12874-018-0482-1 2018

  12. [12]

    gplearn-readthedocs-io-en-stable, 2026

    juan lopez. gplearn-readthedocs-io-en-stable, 2026. URLhttps://www.scribd.com/document/691097226/ gplearn-readthedocs-io-en-stable

    work page arXiv 2026

  13. [13]

    G., Naranjo, S., Rideout, W

    Marco S. Nobile, Paolo Cazzaniga, Daniela Besozzi, Riccardo Colombo, Gi- ancarlo Mauri, and Gabriella Pasi. Fuzzy self-tuning pso: A settings-free algorithm for global optimization.Swarm and Evolutionary Computation, 39:70–85, Apr 2018.https://doi.org/https://doi.org/10.1016/j. swevo.2017.09.001. URLhttps://www.sciencedirect.com/science/ article/pii/S2210...

    work page doi:10.1016/j 2018

  14. [14]

    Palella, Kathleen M

    Frank J. Palella, Kathleen M. Delaney, Anne C. Moorman, Mark O. Loveless, Jack Fuhrer, Glen A. Satten, Diane J. Aschman, and Scott D. Holmberg. Declining morbidity and mortality among patients with ad- vanced human immunodeficiency virus infection.New England Journal of Medicine, 338(13):853–860, Mar 1998.https://doi.org/https://doi. org/10.1056/nejm19980...

    work page doi:10.1056/nejm199803263381301 1998

  15. [15]

    Liah Rosenfeld, Davide Farinati, Diogo Rasteiro, Gloria Pietropolli, Karina Rebuli, Sara Silva, and Leonardo Vanneschi. Slim_gsgp: A python library for non-bloating gsgp.Proceedings of the Genetic and Evolutionary Com- putation Conference, page 1026–1034, Jul 2025.https://doi.org/https: //doi.org/10.1145/3712256.3726398

    work page doi:10.1145/3712256.3726398 2025

  16. [16]

    Hogg, Sharada P

    Hasina Samji, Angela Cescon, Robert S. Hogg, Sharada P. Modur, Keri N. Althoff, Kate Buchacz, Ann N. Burchell, Mardge Cohen, Kelly A. Gebo, M. John Gill, Amy Justice, Gregory Kirk, Marina B. Klein, P. Todd Ko- rthuis, Jeff Martin, Sonia Napravnik, Sean B. Rourke, Timothy R. Sterling, Michael J. Silverberg, and Stephen Deeks. Closing the gap: Increases in ...

    work page doi:10.1371/journal.pone.0081355 2013

  17. [17]

    Development of an algorithm for the charl- son comorbidity index using the national database study in the japan kanto 7 prefectures

    Airi Sekine and Kei Nakajima. Development of an algorithm for the charl- son comorbidity index using the national database study in the japan kanto 7 prefectures. Apr 2025.https://doi.org/https://doi.org/10. 1101/2025.04.10.25325568. URLhttps://www.medrxiv.org/content/ 10.1101/2025.04.10.25325568v1

    work page doi:10.1101/2025.04.10.25325568v1 2025

  18. [18]

    Siegel, Kimberly D

    Rebecca L. Siegel, Kimberly D. Miller, Nikita Sandeep Wagle, and Ahmedin Jemal. Cancer statistics, 2023.CA: A Cancer Journal for Clinicians, 73 (1):17–48, Jan 2023.https://doi.org/https://doi.org/10.3322/caac. 21763. URLhttps://pubmed.ncbi.nlm.nih.gov/36633525/

    work page doi:10.3322/caac 2023

  19. [19]

    Leonardo Vanneschi, Davide Farinati, Diogo Rasteiro, Liah Rosenfeld, Gloria Pietropolli, and Sara Silva. Exploring non-bloating geometric semantic genetic programming.Genetic and Evolutionary Computa- tion, page 237–258, 2025.https://doi.org/https://doi.org/10.1007/ 978-981-96-0077-9_12. URLhttps://link.springer.com/chapter/ 10.1007/978-981-96-0077-9_12

    work page doi:10.1007/978-981-96-0077-9_12 2025