Background

This is a report focusing on the analysis of notes containing phrases associated with end of life decisions and their implementation by the clinical care team. In particular, we are focusing on the National Quality Forum Measure #1626:

Percentage of vulnerable adults admitted to ICU who survive at least 48 hours who have their care preferences documented within 48 hours OR documentation as to why this was not done.



Note: This work will draw heavily from:

  1. Agresti, A. (2013). Categorical Data Analysis, 3rd ed. John Wiley & Sons.
  2. Gelman & Hill (2006). Data Analysis Using Regression and Multilevel/Hierarchical Models, 1st ed. Cambridge University Press.

Data Source

These data were pulled from MIMIC-III, and represent the implementation (as predicted by NeuroNER) of the care measure (between 2008 and 2012) among patients aged > 75 at the time of admission who had Physicians' Notes logged within 48 hours of intake.



Variables of Interest

We are currently looking at the difference in rates of care measure implementation in terms of:

Time Since Admission (Within two days of admission)

Intensive Care Unit on Admission (MICU, SICU, etc.),

Care Provider Expertise (Attending, Resident/Fellow/PA/NP)




Resulting Data

11,104 Observations (Patient Notes)

1075 Patients

1268 Hospital admissions

464 Care providers




Overall Rates of Documentation

Care Provider Level Documentation

##                        
##                            0    1
##   Attending             2232 1314
##   Resident/Fellow/PA/NP 4409 3149
## 
##  Pearson's Chi-squared test with Yates' continuity correction
## 
## data:  test
## X-squared = 21.135, df = 1, p-value = 4.281e-06




ICU Level Documentation

##        
##            0    1
##   CCU    988  555
##   CSRU   290   68
##   MICU  4200 3167
##   SICU   719  416
##   TSICU  444  257




Provider Social Network Graph




Hierarchical (Mixed-Effects) Modeling

Hierarchical models can be helpful in discovering the most important associations given their use of random effects and fixed effects. Because our outcome is binary, we will utilize a Generalized Linear Mixed Effects Model (GLMER).

We are considering the use of three fixed effects:

Time Since Admission, TIME_SINCE_ADMIT (discrete, 0 through 2 days)

First Intensive Careunit, FIRST_CAREUNIT (categorical, MICU, SICU, TSICU, CCU, CSRU)

Care Provider Description, CG_DESCRIPTION (categorical, Attending, Resident/Fellow/PA/NP)

We are considering the care providers, or CGID, as a random effect.


Formally:

\[\overbrace{\mathbf{y}}^{\mbox{1075 x 1}} \quad = \quad \overbrace{\underbrace{\mathbf{X}}_{\mbox{1075 x 4}} \quad \underbrace{\boldsymbol{\beta}}_{\mbox{Fixed Effect Coefficients}}}^{\mbox{1075 x 1}} \quad + \quad \overbrace{\underbrace{\mathbf{Z}}_{\mbox{1075 x 464}} \quad \underbrace{\boldsymbol{u}}_{\mbox{Random Effects}}}^{\mbox{1075 x 1}} \quad + \quad \overbrace{\boldsymbol{\varepsilon}}^{\mbox{1075 x 1}}\]


Care Providers As Random Effects (Utilizing Partial Pooling)




Due to variation between the amount of patients seen by Attending physicians as compared to Residents over time, we will treat clinicians themselves as random effects, which implies the use of partial pooling within the model. We can demonstrate this by randomly sampling 30 care providers and modeling their documentation over the first two days of patient admission.




## 
## CG_14010 CG_14022 CG_14037 CG_14045 CG_14056 CG_14080 CG_14110 CG_14114 
##       13        1       94       37       13        6        9        7 
## CG_14116 CG_14126 CG_14154 CG_14180 CG_14184 CG_14240 CG_14252 CG_14269 
##       62        6       34       45        2       20       77       41 
## CG_14307 CG_14330 CG_14331 CG_14354 CG_14356 CG_14384 CG_14393 CG_14417 
##        3       48        2        2        9        6       15       20 
## CG_14444 
##       16







Preliminary Results (Model Output Summary)

## Generalized linear mixed model fit by maximum likelihood (Adaptive
##   Gauss-Hermite Quadrature, nAGQ = 10) [glmerMod]
##  Family: binomial  ( logit )
## Formula: Caremeasure_Implementation ~ TIME_SINCE_ADMIT + FIRST_CAREUNIT +  
##     CG_DESCRIPTION + (1 | CGID)
##    Data: dat
## Control: glmerControl(optimizer = "bobyqa")
## 
##      AIC      BIC   logLik deviance df.resid 
##  14215.0  14273.5  -7099.5  14199.0    11096 
## 
## Scaled residuals: 
##     Min      1Q  Median      3Q     Max 
## -2.2801 -0.7880 -0.5058  0.9837  3.5087 
## 
## Random effects:
##  Groups Name        Variance Std.Dev.
##  CGID   (Intercept) 0.8456   0.9196  
## Number of obs: 11104, groups:  CGID, 464
## 
## Fixed effects:
##                                     Estimate Std. Error z value Pr(>|z|)
## (Intercept)                         -0.79524    0.14836  -5.360 8.31e-08
## TIME_SINCE_ADMIT                    -0.07629    0.03181  -2.398   0.0165
## FIRST_CAREUNITCSRU                  -0.34878    0.16977  -2.054   0.0399
## FIRST_CAREUNITMICU                   0.33219    0.07358   4.515 6.34e-06
## FIRST_CAREUNITSICU                   0.42726    0.10691   3.997 6.43e-05
## FIRST_CAREUNITTSICU                  0.58015    0.12860   4.511 6.44e-06
## CG_DESCRIPTIONResident/Fellow/PA/NP  0.08846    0.14334   0.617   0.5372
##                                        
## (Intercept)                         ***
## TIME_SINCE_ADMIT                    *  
## FIRST_CAREUNITCSRU                  *  
## FIRST_CAREUNITMICU                  ***
## FIRST_CAREUNITSICU                  ***
## FIRST_CAREUNITTSICU                 ***
## CG_DESCRIPTIONResident/Fellow/PA/NP    
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Correlation of Fixed Effects:
##                 (Intr) TIME_S FIRST_CAREUNITC FIRST_CAREUNITM
## TIME_SINCE_     -0.261                                       
## FIRST_CAREUNITC -0.174 -0.014                                
## FIRST_CAREUNITM -0.362 -0.008  0.296                         
## FIRST_CAREUNITS -0.297  0.005  0.273           0.564         
## FIRST_CAREUNITT -0.271  0.006  0.251           0.469         
## CG_DESCRIPT     -0.795  0.012  0.003          -0.036         
##                 FIRST_CAREUNITS FIRST_CAREUNITT
## TIME_SINCE_                                    
## FIRST_CAREUNITC                                
## FIRST_CAREUNITM                                
## FIRST_CAREUNITS                                
## FIRST_CAREUNITT  0.529                         
## CG_DESCRIPT     -0.026          -0.014




Model Results (Odds Ratios, 95% Confidence Intervals)

    Caremeasure_Implementation
    Odds Ratio CI p
Fixed Parts
(Intercept)   0.45 0.34 – 0.60 <.001
TIME_SINCE_ADMIT   0.93 0.87 – 0.99 .016
FIRST_CAREUNIT (CSRU)   0.71 0.51 – 0.98 .040
FIRST_CAREUNIT (MICU)   1.39 1.21 – 1.61 <.001
FIRST_CAREUNIT (SICU)   1.53 1.24 – 1.89 <.001
FIRST_CAREUNIT (TSICU)   1.79 1.39 – 2.30 <.001
CG_DESCRIPTION (Resident/Fellow/PA/NP)   1.09 0.82 – 1.45 .537
Random Parts
τ00, CGID   0.846
NCGID   464
ICCCGID   0.204
Observations   11104
Deviance   13296.641




Fixed Effects Results