From data to information
Robust Logistic Modeling and transference to an App in
estimating the risk of Stroke mortality in the emergency
department
Communication 3
Juan Manuel García-Torrecillas. Hospital Torrecárdenas, Almería
Jesús de la Fuente Arias. University of Almeria
Giulliana Solinas. University of Sassary
Gabrielle Giorgy. University of Roma
University of
Almería
Stroke. Epidemiology Notes
From: Serrano Castro, PJ. Samfyc 2014
Third etiology of death in Occident
USA: 1 event every 40 seconds
Incidence (Europe): 95-290 ep/100.00 p-year
High rate of complications and sequels
Mortality Rate 36 episodes/100,000person-year
First etiology of death in Spanish women
Is an time-dependent disease
Importance of
individual and
contextual
factors
Importance of
specialted
personal in the
first atention
What about benefits of knowing Stroke´s risk of mortality?
PRINCIPAL
To detect associated factors to in
hospital mortality in patients
admited by isquémic stroke
ESPECIFIC
To evaluate the Discriminative
ability (trough de AUC) and
Callibration (trough visual risk
centiles of Hosmer Lemeshow
Test)
All the episodes of DRG 14 admited in Spanish hospitals during the period
2008-2014
Source Information
CMBDH years 2008-14
-Heath and Social Politics Ministery
-Statistical Web of Health Ministery
- National Health Statistics
Diagnosis Codification
CIE 9- MC
AP-GRD versión 21.0
Criterios de Urrea et al
GRD 14 (Stroke)
2008-2012
186,245 episodes
Observational anality study.
Historic Cohort, DRG 14 (Ischemic Stroke)
Design and Episodes in the study
Explanatory Variables
In Hospital Mortality
Dependent Variable
Variables in the study
Age (years)
Year (2008-2012)
Sex (male-female)
AACC (Méth. Foster/SMR)
Brain topography
Clinical Comorbidities
Financials
Dates: Born-Admission-Discharge
Type of Admission: (UI-PI)/ Type of
discharge
CIE Diagnostic (14 places)
CIE Procedures (20 places)
Mortality
Clinical Complications
Descriptive and Inference
Inferential Study
Descriptive
Exploratory
Cuantitative Variables
Means (SD), CI 95% normal method
Categórical Variables
Frecuency Tables; % & CI 95%
Cualitative Variables
t de Student (con Levene)
Categorical Variables
Pearson χ2
CI Examination
Bivariant Studies
OR (CI 95%)
Robust Logistic Regression
OR adjusted (CI 95%)
Discrimination: ROC. C-Statistic
Calibration: Hosmer-Lemeshow
Predictive Model
Constant
observation of
the non-
differential
classification
bias
Descriptive and Evolution of Admissions
186,245 patients admitted (episodes)
Length of stay: 7,54 (4,54) days
Age: 73,92(12,54) years
NDA: 6,91 (2.96) diag.
NPA: 3,27 (2,45) proc.
Readmissions: 4,8%
Exitus: 6,9%
36900
37000
37100
37200
37300
37400
37500
2008_
2009_
2010_
2011_
2012_Año
37084
37114
37399
37228
37420
Admissions/Year
Avoiding Overfitting with Decission Trees
Robust Logistic Regression
Method of logistic: Forced Enter
Resampling (Bootstraping)
APPs for Ictus
- Quick
stimation of
risk
- Agile
- Easy

Robust Logistic Modeling and transference to an App in estimating the risk of Stroke mortality in the emergency department

  • 1.
    From data toinformation Robust Logistic Modeling and transference to an App in estimating the risk of Stroke mortality in the emergency department Communication 3 Juan Manuel García-Torrecillas. Hospital Torrecárdenas, Almería Jesús de la Fuente Arias. University of Almeria Giulliana Solinas. University of Sassary Gabrielle Giorgy. University of Roma University of Almería
  • 2.
    Stroke. Epidemiology Notes From:Serrano Castro, PJ. Samfyc 2014
  • 3.
    Third etiology ofdeath in Occident USA: 1 event every 40 seconds Incidence (Europe): 95-290 ep/100.00 p-year High rate of complications and sequels Mortality Rate 36 episodes/100,000person-year First etiology of death in Spanish women Is an time-dependent disease Importance of individual and contextual factors Importance of specialted personal in the first atention What about benefits of knowing Stroke´s risk of mortality?
  • 4.
    PRINCIPAL To detect associatedfactors to in hospital mortality in patients admited by isquémic stroke ESPECIFIC To evaluate the Discriminative ability (trough de AUC) and Callibration (trough visual risk centiles of Hosmer Lemeshow Test)
  • 6.
    All the episodesof DRG 14 admited in Spanish hospitals during the period 2008-2014 Source Information CMBDH years 2008-14 -Heath and Social Politics Ministery -Statistical Web of Health Ministery - National Health Statistics Diagnosis Codification CIE 9- MC AP-GRD versión 21.0 Criterios de Urrea et al GRD 14 (Stroke) 2008-2012 186,245 episodes Observational anality study. Historic Cohort, DRG 14 (Ischemic Stroke) Design and Episodes in the study
  • 7.
    Explanatory Variables In HospitalMortality Dependent Variable Variables in the study Age (years) Year (2008-2012) Sex (male-female) AACC (Méth. Foster/SMR) Brain topography Clinical Comorbidities Financials Dates: Born-Admission-Discharge Type of Admission: (UI-PI)/ Type of discharge CIE Diagnostic (14 places) CIE Procedures (20 places) Mortality Clinical Complications
  • 8.
    Descriptive and Inference InferentialStudy Descriptive Exploratory Cuantitative Variables Means (SD), CI 95% normal method Categórical Variables Frecuency Tables; % & CI 95% Cualitative Variables t de Student (con Levene) Categorical Variables Pearson χ2 CI Examination Bivariant Studies OR (CI 95%) Robust Logistic Regression OR adjusted (CI 95%) Discrimination: ROC. C-Statistic Calibration: Hosmer-Lemeshow Predictive Model Constant observation of the non- differential classification bias
  • 10.
    Descriptive and Evolutionof Admissions 186,245 patients admitted (episodes) Length of stay: 7,54 (4,54) days Age: 73,92(12,54) years NDA: 6,91 (2.96) diag. NPA: 3,27 (2,45) proc. Readmissions: 4,8% Exitus: 6,9% 36900 37000 37100 37200 37300 37400 37500 2008_ 2009_ 2010_ 2011_ 2012_Año 37084 37114 37399 37228 37420 Admissions/Year
  • 11.
  • 12.
    Robust Logistic Regression Methodof logistic: Forced Enter Resampling (Bootstraping)
  • 14.
    APPs for Ictus -Quick stimation of risk - Agile - Easy