November 9, 2018

Format

  • Introduction

  • Methods

  • Results

  • Discussion

  • Reproducibility

Introduction

Gap in literature

The Burden of Traumatic Brain Injury (TBI)

  • 69 million cases annually

  • Disproportionately in LMICs

  • Timely surgical intervention can prevent disability or death

  • Impact of surgery on outcomes in LMICs largely unexplored

Research Objective

Determine if TBI patients receiving surgery have better outcomes compared to TBI patients not receiving surgery and how or if that differs with severity of injury.

Methods

Methods

Data

  • KCMC TBI registry

  • May 2014 to April 2017

Survival Analysis

  • Kaplan Meier (KM) Plots

  • Cox proportional hazard model

type variable Definition
explanatory Adm Date Date of admission
Outcome Date Date or outcome, either hospital discharge or inpatient death
MOI Mechanism of Injury: Fall, RTI, Assault, Other
GCS Classic GCS: 14-15-Mild, 9-13-Moderate, 3-8-Severe
Age All ages included
Gender Male / Female
ICU If patient was transferred from ward to ICU
EtOH Was alcohol involved at time of injury
TBI Surgery Includes mostly burrholes, some craniotomies/ectomies
outcome GOS Scale from 1-5, 1-3 = Poor outcome, 4-5 = Good outcome

Analysis

KM Plots and Cox Proportional Hazard Model

Survival Analysis

KM Plots -

Cox Model -

Survival Analysis

KM Plots - Visualize events over time

Cox Model -

Survival Analysis

KM Plots - Visualize events over time

Cox Model -

Survival Analysis

KM Plots - Visualize events over time

Cox Model - Quantifying the effect of a treatment on an outcome (Hazard Ratio)

Survival Analysis

KM Plots - Visualize events over time

Cox Model - Quantifying the effect of a treatment on an outcome (Hazard Ratio)

Results

Table One

n = 2506
n = 628
n = 1878
variable value Yes TBI Surgery No TBI Surgery
Sociodemographic
<18 (n%) 369 ( 15 ) 100 ( 16 ) 269 ( 15)
18-29 (n%) 883 ( 36 ) 202 ( 32 ) 681 ( 37)
30-39 (n%) 559 ( 23 ) 147 ( 24 ) 412 ( 22)
40-49 (n%) 325 ( 13 ) 88 ( 14 ) 237 ( 13)
50+ (n%) 343 ( 14 ) 88 ( 14 ) 255 ( 14)
Male (n%) 2087 ( 83 ) 526 ( 84 ) 1561 ( 83)
EtOH
Yes (n%) 667 ( 27 ) 153 ( 24 ) 514 ( 28)
No (n%) 1218 ( 49 ) 310 ( 50 ) 908 ( 49)
Unknown (n%) 599 ( 24 ) 162 ( 26 ) 437 ( 24)
Mechanism of Inj
RTI (n%) 1695 ( 75 ) 349 ( 63 ) 1346 ( 79)
Violence (n%) 362 ( 16 ) 127 ( 23 ) 235 ( 14)
Fall (n%) 252 ( 11 ) 72 ( 13 ) 180 ( 11)
Other (n%) 189 ( 8 ) 75 ( 14 ) 114 ( 7 )
Clinical Data
Yes_icu (n%) 367 ( 15 ) 240 ( 41 ) 127 ( 7 )
Bad Recovery (n%) 257 ( 10 ) 90 ( 14 ) 167 ( 9 )
GCS Total Mean (SD) 13.2 ( 3.2) 12.4 ( 3.7) 13.5 ( 3 )
Mild (n%) 1833 ( 73 ) 388 ( 62 ) 1445 ( 77)
Moderate (n%) 357 ( 14 ) 117 ( 19 ) 240 ( 13)
Severe (n%) 316 ( 13 ) 123 ( 20 ) 193 ( 10)

Results

KM Plots

Results

Cox Model

Model Output

  • Ran the model 4 times

  • Results are controlling for all covariates

  • Only showing TBI surgery output

  • 1 to 2 time-interaction terms needed

  • Hazard ratio (HR)

  • Less than one is good

Cox Model Output: HR (CI)
HD Overall Mild Moderate Severe
1 .32
(.19, .54) 		.20
(.06, .64) 		.17
(.06, .49) 		.47
(.24, .89)
2
3
4 .46
(.25, .86) 		.65
(.30, 1.39)
5
6
7
8 .77
(.36, 1.66) 		.69
(.14, 3.31) 		1.23
(.23, 6.68) 		.70
(.23, 2.13)
9
10
11
12
13
14

Model Output: Entire study with all covariates

  Decrease chance of poor outcome

  • Surgery and outcome before HD 8

 

Increase chance of poor outcome

  • ICU

  • Mod / severe TBI

  • Age

Cox Model Output: HR (CI)
Variable HR_CI Pvalue
Gender
Male 1.51 (0.98, 2.35) 0.062
Age
Age_less_18 ref NA
Age_18_29 1.91 (1.04, 3.52) 0.037
Age_30_39 2.45 (1.29, 4.64) 0.006
Age_40_49 2.51 (1.28, 4.91) 0.007
Age_50 1.92 (0.99, 3.72) 0.053
Mechanism of Inj
RTI ref NA
Assault 0.73 (0.41, 1.32) 0.302
Fall 1.55 (0.96, 2.51) 0.072
Other 1.24 (0.67, 2.28) 0.493
EtOH
No ref NA
Yes 0.90 (0.59, 1.38) 0.633
Unknown 1.03 (0.70, 1.52) 0.878
TBI Surgery
HD_1_3 0.32 (0.19, 0.54) 0.001
HD_4_7 0.46 (0.25, 0.86) 0.014
HD_8_14 0.77 (0.36, 1.66) 0.505
ICU Transfer
Yes_ICU 3.43 (2.32, 5.05) 0.001
GCS
Mild ref NA
Moderate 3.21 (1.99, 5.17) 0.001
Severe 8.01 (5.18, 12.4) 0.001

Discussion

Discussion

First analysis of acute TBI outcome using survival analysis in LMIC

 

One of the largest single-center studies on head injury

 

Surgery beneficial for all severities with early outcome

 

The severity-dependent influence of surgery on outcomes

  • Moderates (HR: .17) > Mild (HR: .20) > Severe (HR: .47)

Discussion cont.

Benefit of surgery on outcomes decreases for those with outcome after HD 7

Cox Model Output: HR (CI)
Variable HR_CI Pvalue
TBI Surgery
HD_1_3 0.32 (0.19, 0.54) 0.001
HD_4_7 0.46 (0.25, 0.86) 0.014
HD_8_14 0.77 (0.36, 1.66) 0.505

  • Possible delay to diagnosis and treatment

  • Investigate post-operation care for complex surgery patients

Interesting finding

  • Severity dependent curves

  • Interaction terms crucial

  • Moderate curve is unique

Reproducibility

Reproducibility

Methods

Patient population

The study included patients enrolled in the KCMC registry between start_date and end_date. We excluded patients if they had missing data in the following key variables: admission date, discharge date, received surgery for TBI, admission Glasgow coma scale (GCS) and discharge Glasgow outcome scale (GOS).

#Loading in your data file. Set the working directory to where the data is saved on your computer
setwd("C:/Users/cyrus/OneDrive/Documents/R/input")
#Insert the name of your data file to read in the data
tbi <- read.csv("clean_tbi.csv")

#This processess the dates for arrival and outcome
tbi$date_arrival2 <- (strptime  ( paste(tbi$date_arrival, tbi$time_arrival), "%m/%d/%Y %H:%M"))
tbi$date_death2   <- (strptime  ( paste(tbi$date_death, tbi$time_death), "%m/%d/%Y %H:%M") )
tbi$date_home2    <- (strptime  ( paste(tbi$date_dc_home, tbi$time_home), "%m/%d/%Y %H:%M"))

#Start and end dates of your data are stored here
start_date <- min(test6$date_arrival2,na.rm = T)
end_date <- max(test6$date_arrival2,na.rm = T)

Reproducibility

Weebale

(Luganda for thank you)