Skip to content

Appendix.md

Latest commit

 

History

History
48 lines (26 loc) · 7.33 KB

Appendix.md

File metadata and controls

48 lines (26 loc) · 7.33 KB

We provide additional experimental details and results including the KM plots in Kaplan-Meier (KM) plots, the imprtant features identified by all methods for Breast Cancer, SUPPORT, FLCHAIN, and EHR datasets in Important Features, and some of the feature calibration functions obtained using IRIS on SUPPORT, FLCHAIN, AV45, and EHR datasets in Feature Calibration Functions.

KM Plots

The Kaplan–Meier (KM) plots in the folder Kaplan-Meier (KM) plots correspond to the results presented in TABLE II. Each plot illustrates the survival estimates for two identified clusters: the blue curve represents higher survival probability, and the orange curve represents lower survival probability.

Across all datasets, IRIS consistently produces the most distinct separation between survival curves, as reflected in the highest LogRank values, demonstrating its effectiveness in stratifying individuals into high- and low-risk groups.

  • In BreastCancer plots, Deep Cox and CoxNAM perform similarly, while DCSM shows a weaker separation.
  • In SUPPORT plots, both DCSM and CoxNAM achieve strong separation, while Deep Cox exhibits the lowest separation, indicating weaker risk stratification.
  • In FLCHAIN plots, DCSM performs comparably to IRIS, whereas CoxNAM results in poor clustering.
  • In EHR plots, DCSM shows better performance than Deep Cox and CoxNAM.

Important Features and Feature Calibration Functions

Tables in Important Features present the key features extracted by various methods, corresponding to the results in Table III.

IRIS identifies creatinine, monoclonal gammopathy of undetermined significance (MGUS), age, lambda, and kappa as the most significant features from FLCHAIN dataset. As shown in Figures of FLCHAIN Feature Calibration, the feature calibration functions illustrate the relationships between these variables and mortality risk. Notably, elevated serum creatinine levels at admission are strongly associated with higher mortality rates [1], and increased serum free light chains (sFLC), particularly kappa and lambda, are linked to increased mortality risk in individuals with chronic kidney disease (CKD) [2].

Table SUPPORT Feature Calibration highlights level of functional disability (sfdm2) as a crucial predictor identified by IRIS. This feature, after one-hot encoding, maintains high importance across all values. Patients classified under "<2-month follow-up" are at the highest severity level, as shown in [Link](feature_calibration_functions/SUPPORT/fim_plot_seed_1009_k0_support_IRIS_sfdm2_<2 mo. follow-up.pdf). For high-risk patients, both decreased respiratory rate [3] and low bilirubin levels [4] are linked to worsening health outcomes, as reflected in Link and Link, respectively.

Table EHR Dataset Feature Importance indicates that IRIS identifies several Social Determinants of Health (SDOH) as important features, including environment, alcohol use, psychoactive substance use, and tobacco use. Environmental factors play a critical role in Alzheimer’s Disease and Related Dementias (ADRD) [[5]], and broader indicators like neighborhood deprivation have also been associated with ADRD risk [[6]].

Among behavioral factors:

  • Heavy alcohol consumption is linked to accelerated cognitive decline in Alzheimer’s patients [3].
  • Psychoactive substance use has been associated with increased dementia risk and structural changes in the aging brain [7].

IRIS effectively captures these relationships, highlighting the connection between alcohol use, psychoactive substances, and injury risk, as shown in Link and Link. Additionally, smoking increases harmful brain stress, contributing to Alzheimer’s disease and memory loss [8]. Older adults with a history of smoking often show brain changes resembling those seen in Alzheimer’s, reflecting long-term neurobiological effects. The shape function corresponding to this is shown in Link.

References

  1. Mehmet Akif Cakar et al., The Effect of Admission Creatinine Levels on One-Year Mortality in Acute Myocardial Infarction, The Scientific World Journal, 2012. Link
  2. Fraser, S. D. S., Fenton, A., Harris, S., et al. (2017). The Association of Serum Free Light Chains with Mortality and Progression to End-Stage Renal Disease in Chronic Kidney Disease: Systematic Review and Individual Patient Data Meta-analysis. Mayo Clinic Proceedings, 92(11), 1671–1681. Link
  3. Strauß, R., Ewig, S., Richter, K., König, T., Heller, G., & Bauer, T. T. (2014). The Prognostic Significance of Respiratory Rate in Patients with Pneumonia: A Retrospective Analysis of Data from 705,928 Hospitalized Patients in Germany from 2010–2012. Dtsch Arztebl Int, 111(29–30), 503–508. https://doi.org/10.3238/arztebl.2014.0503
  4. Ong, K. L., Allison, M. A., Cheung, B. M., Wu, B. J., Barter, P. J., & Rye, K. A. (2014). The Relationship Between Total Bilirubin Levels and Total Mortality in Older Adults: The United States National Health and Nutrition Examination Survey (NHANES) 1999–2004. PLoS One, 9(4), e94479. https://doi.org/10.1371/journal.pone.0094479
  5. Adkins-Jackson, P. B., George, K. M., Besser, L. M., Hyun, J., Lamar, M., Hill-Jarrett, T. G., Bubu, O. M., Flatt, J. D., Heyn, P. C., Cicero, E. C., et al. (2023). The structural and social determinants of Alzheimer's disease related dementias. Alzheimer's & Dementia, 19(7), 3171–3185.
  6. Powell, W. R., Buckingham, W. R., Larson, J. L., Vilen, L., Yu, M., Salamat, M. S., Bendlin, B. B., Rissman, R. A., & Kind, A. J. H. (2020). Association of Neighborhood-Level Disadvantage With Alzheimer Disease Neuropathology. JAMA Network Open, 3(6), e207559.
  7. Tournier, M., Pambrun, E., Maumus-Robert, S., Pariente, A., & Verdoux, H. (2022). The risk of dementia in patients using psychotropic drugs: antidepressants, mood stabilizers or antipsychotics. Acta Psychiatrica Scandinavica, 145(1), 56–66. Link
  8. Durazzo, T. C., Mattsson, N., & Weiner, M. W. (2014). Smoking and increased Alzheimer's disease risk: a review of potential mechanisms. Alzheimer's & Dementia, 10, S122–S145. Link