Comparison of Artificial Neural Networks and Logistic Regression for 30-days Survival Prediction of Cancer Patients

Introduction

A machine learning technique that imitates neural system and brain can provide better than traditional methods like logistic regression for survival prediction and create an algorithm by determining influential factors.

Aim

To determine the influential factors on survival time of palliative care cancer patients and to compare two statistical methods for better prediction of survival.

Methods

One-year data is gathered from the patients that we followed in the palliative care clinic of our hospital (2017-2018) (n = 189). All data were retrospectively evaluated. After descriptive statistics, we used Pearson and Spearman correlations for parametric and non-parametric variables. The Artificial Neural Networks (ANN) and logistic regression model were applied to parameters which have a significant correlation with short survival.

Results

Significantly correlated variables with short survival were Palliative Performance Scale (PPS), Edmonton Symptom Assessment System (ESAS), Karnofsky Performance Scale (KPS), brain, liver, and distant metastasis, hemogram parameters, cero-reactive protein (CRP) and albumin (ALB). ANN model showed 89.3% prediction accuracy while the logistic regression model showed 73.0%. ANN model achieved a better AUC value of 0.86 than logistic regression model (0.76).

Discussion

There are several prognostic evaluation tools such as PPS, KPS, CRP, albumin, leukocytes, neutrophil were reported several studies as survival-related parameters in logistic regression models, also. Many studies compare ANN with logistic regression. When we evaluated these parameters totally, we observed the same relations with survival then we used the same parameters in the ANN model. The effectivity of the survival prediction models can be improved with the use of ANN.

Conclusion

ANN provides a more accurate estimation than logistic regression. ANN model is an important statistical method for survival prediction of cancer patients.

Patient and caregiver agreement on prognosis estimates for older adults with advanced cancer

BACKGROUND

Disagreements between patients and caregivers about treatment benefits, care decisions, and patients' health are associated with increased patient depression as well as increased caregiver anxiety, distress, depression, and burden. Understanding the factors associated with disagreement may inform interventions to improve the aforementioned outcomes.

METHODS

For this analysis, baseline data were obtained from a cluster-randomized geriatric assessment trial that recruited patients aged ≥70 years who had incurable cancer from community oncology practices (University of Rochester Cancer Center 13070; Supriya G. Mohile, principal investigator). Patient and caregiver dyads were asked to estimate the patient's prognosis. Response options were 0 to 6 months, 7 to 12 months, 1 to 2 years, 2 to 5 years, and >5 years. The dependent variable was categorized as exact agreement (reference), patient-reported longer estimate, or caregiver-reported longer estimate. The authors used generalized estimating equations with multinomial distribution to examine the factors associated with patient-caregiver prognostic estimates. Independent variables were selected using the purposeful selection method.

RESULTS

Among 354 dyads (89% of screened patients were enrolled), 26% and 22% of patients and caregivers, respectively, reported a longer estimate. Compared with dyads that were in agreement, patients were more likely to report a longer estimate when they screened positive for polypharmacy (β = 0.81; P = .001), and caregivers reported greater distress (β = 0.12; P = .03). Compared with dyads that were in agreement, caregivers were more likely to report a longer estimate when patients screened positive for polypharmacy (β = 0.82; P = .005) and had lower perceived self-efficacy in interacting with physicians (β = -0.10; P = .008).

CONCLUSIONS

Several patient and caregiver factors were associated with patient-caregiver disagreement about prognostic estimates. Future studies should examine the effects of prognostic disagreement on patient and caregiver outcomes.

Does Calculated Prognostic Estimation Lead to Different Outcomes Compared With Experience-Based Prognostication in the ICU? A Systematic Review

Little is known about the impact of providing calculator/guideline based versus clinical experiential-based prognostic estimates to patients/caregivers in the ICU. We sought to determine whether studies have compared types of prognostic estimation in the ICU and associations with outcomes.

Data Sources

Following Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines, databases searched were PubMed, Embase, Web of Science, and Cochrane Library. The search was run on January 4, 2016, and April 12, 2017. References for included articles were searched.

Study Selection

Studies meeting the following criteria were included in the analysis: communication of prognostic estimates, a comparator group, and in the adult ICU setting.

Data Extraction

Titles/abstracts were reviewed by two researchers. We identified 10,704 articles of which 10 met inclusion criteria. Seven of the studies included estimates obtained from calculators/guidelines and three were based on subjective estimation wherein clinicians were asked to estimate prognosis based on experience. Only the seven using calculated/guideline based estimation were used for pooled analysis. Of these, one was a randomized trial, and six were nonrandomized before/after studies. All of the studies communicated the calculated/guideline-based estimates to the clinician. Two studies involved the communication of calculated prognostic estimates to the ICU physicians for all ICU patients. Four included identification of high-risk patients based on guidelines or review of historical local data which triggered a palliative care/ethics consultation, and one study included communication to physicians about guideline based likely outcomes for neurologic recovery for patients with out-of-hospital cardiac arrest survivors. The comparator arm in all studies was usual care without protocolized prognostication.

Data Synthesis

Included studies were assessed for risk of bias. The most common outcomes measured were hospital mortality; do-not-resuscitate status; and medical ICU length of stay. In pooled analyses, there was an association between calculated/guideline based prognostic estimation and decreased medical ICU length of stay as well as increased do-not-resuscitate status, but no difference in hospital mortality.

Conclusions

Protocolized assessment of calculator/guideline based prognosis in ICU patients is associated with decreased medical ICU length of stay and increased do-not-resuscitate status but does not have a significant effect on mortality. Future studies should explore how communicating these estimates to physicians changes behaviors including communication to patients/families and whether calculator/guideline based prognostication is associated with improved patient and family rated outcomes.