A MACHINE LEARNING-PREDICTIVE MODEL TO IDENTIFY FACTORS ASSOCIATED WITH DEATH IN DIALYSIS PATIENTS

https://storage.unitedwebnetwork.com/files/1099/0822fff3620b33db8bb54dacfd0e24e7.pdf
A MACHINE LEARNING-PREDICTIVE MODEL TO IDENTIFY FACTORS ASSOCIATED WITH DEATH IN DIALYSIS PATIENTS
Daniela
Ponce
Luis Gustavo Modelli de Andrade gustavo.modelli@unesp.br UNESP Departamento de Clínica Médica Botucatu
Carolina Aparecida de Almeida Vicentini carolina.aparecida@unesp.br Faculdade de Medicina de Botucatu Faculdade de Medicina de Botucatu Botucatu
 
 
 
 
 
 
 
 
 
 
 
 
 

Chronic Kidney Disease (CKD) is a significant global health concern, and dialysis patients have a higher mortality rate from cardiovascular disease (CVD) compared to the general population. The use of Machine Learning (ML) and Artificial Intelligence in medicine has grown exponentially over the past few decades. However, few studies have focused on predictive models for determining factors associated with death in dialysis patients, with a predominant focus on the hemodialysis (HD) population. The primary objective of this study was to identify mortality-related factors in patients undergoing both hemodialysis (HD) and peritoneal dialysis (PD), encompassing both planned and urgent starts. 

R software algorithms were employed to develop ML predictive models. The study included adult patients undergoing HD and PD, either in a planned or urgent manner, between January 2014 and January 2019, at Hospital das Clínicas da Faculdade de Medicina de Botucatu located in Botucatu, São Paulo, Brazil. Epidemiological, clinical, and laboratory data were collected. 

The results showed that death occurred in 170 patients (29.3%). Cox regression analysis revealed that death was associated with older age, fewer Exit Site Infection(ESI)-free months, lower initial creatinine, dialysis-related infection (peritonitis for PD and bloodstream infection for HD), and hospitalizations (Table 1). Random forest (Figure 1) ranked the following main variables predictive of death in descending order of importance: ESI-free months; age and initial levels of creatinine. 


Table 01 – COX Regression based on Machine Learning

Variables

HR1

95% CI1

p

Age

1.02

1.01- 1.04

<0.001

Male gender

1.07

0.75- 1.52

0.7

Number of comorbidities

1.08

0.95- 1.21

0.2

(ESI-free days)/(30)*

0.96

0.94- 0.98

<0.001

Dialysis-related infection

0.55

0.37- 0.82

0.003

PTH

1.00

1.00- 1.00

0.2

Hemoglobin

0.93

0.84- 1.03

0.2

Albumin

0.83

0.60- 1.14

0.3

P

0.99

0.96- 1.02

0.6

Diabetes

0.81

0.55- 1.19

0.3

Creatinine

0.90

0.83- 0.98

0.015

Hospitalizations

1.72

1.08- 2.74

0.023

CVC for initial access (HD)

1.84

0.44- 7.71

0.4

APD as the initial modality (PD)

2.33

0.55- 9.90

0.3

AFV for initial dialysis access

1.00

0.21- 4.64

0.9

Dialysis modality switching

0.95

0.55- 1.65

0.9

1HR = Hazard Ratio, CI = Confidence Interval

*ESI-Free months

Figure 1: Variables importance ranked by Random Forest



In conclusion, this study revealed that ESI-free months, age and initial creatinine levels were associated with death on both multivariate and ML-based analyses. The content presented in this abstract was submitted for other meetings.

E-Poster Format Requirements
  • PDF file
  • Layout: Portrait (vertical orientation)
  • One page only (Dim A4: 210 x 297mm or PPT)
  • E-Poster can be prepared in PowerPoint (one (1) PowerPoint slide) but must be saved and submitted as PDF file.
  • File Size: Maximum file size is 2 Megabytes (2 MB)
  • No hyperlinks, animated images, animations, and slide transitions
  • Language: English
  • Include your abstract number
  • E-posters can include QR codes, tables and photos