======================================================================
Logistic Regression on Heart Failure Data with Bootstrap Inference
======================================================================

DATA DESCRIPTION
----------------------------------------------------------------------
Source: UCI Machine Learning Repository
URL: https://archive.ics.uci.edu/dataset/519/heart+failure+clinical+records
Reference: Ahmad et al. (2017), PLoS ONE 12(7):e0181001
Analysis: Chicco and Jurman (2020), BMC Med Inform Decis Mak 20:16
Sample size: n = 299 
Number of deaths: 96 ( 32.1 %)

MODEL
----------------------------------------------------------------------
Response: DEATH_EVENT (0 = survived, 1 = died)
Predictors: Age, Ejection Fraction, Serum Creatinine, Follow-up Time

Model: log(p/(1-p)) = b0 + b1*Age + b2*EF + b3*SC + b4*Time
(Predictors standardized to mean 0, SD 1.)

BOOTSTRAP STRATEGY
----------------------------------------------------------------------
Method:    Pairs/case bootstrap (resample (X, y) rows with replacement)
N:          1000 
Estimator: Newton-Raphson MLE on each bootstrap sample

COEFFICIENT ESTIMATES
----------------------------------------------------------------------
Coefficient            Estimate      Wald SE      Boot SD
----------------------------------------------------------------------
Intercept               -1.2902       0.1953       0.2177
Age                      0.5154       0.1769       0.1700
Ejection Fraction       -0.8853       0.1841       0.2144
Serum Creatinine         0.7446       0.1806       0.2818
Time                    -1.5997       0.2237       0.2621

95% PERCENTILE BOOTSTRAP CIs (PRIMARY)
----------------------------------------------------------------------
Intercept             -1.2902  (95% CI:  -1.8348,  -0.9622)
Age                    0.5154  (95% CI:   0.2361,   0.9002)
Ejection Fraction     -0.8853  (95% CI:  -1.3989,  -0.5393)
Serum Creatinine       0.7446  (95% CI:   0.3310,   1.4493)
Time                  -1.5997  (95% CI:  -2.2525,  -1.2106)

INTERPRETATION (on standardized scale)
----------------------------------------------------------------------
Each coefficient gives the change in log-odds of death per 1-SD
change in the predictor (with the other three held at their means).
Negative coefficients (Ejection Fraction, Time) are protective.
Positive coefficients (Age, Serum Creatinine) increase mortality risk.