Prediction of In-Hospital Mortality Following Vertebral Fracture Fixation in Patients With Ankylosing Spondylitis or Diffuse Idiopathic Skeletal Hyperostosis: Machine Learning Analysis

Andrew Cabrera; Alexander Bouterse; Michael Nelson; Coleman Dietrich; Jacob Razzouk; Udochukwu Oyoyo; Christopher M. Bono; Olumide Danisa

doi:10.14444/8567

Article Figures & Data

Figures

Tables
Supplementary Materials

Download figure
Open in new tab
Download powerpoint
Figure
Graphical representation of each algorithm’s area under the receiver operating characteristics curve (AUROC) in prediction of in-hospital mortality. Abbreviations: ADAboost, Adaptive Boosting Classifier; GB, Gradient Boosting Classifier; RF, Random Forest Classifier.

View popup

Table 1

Patient characteristics (n = 2960).

Characteristic	Mean ± SD or % (n)
Age, y, mean ± SD	73.13 ± 10.10
Sex
Male	84.5% (2500)
Female	15.5% (460)
Race
White	76.0% (2250)
African American	4.6% (135)
Hispanic	8.3% (245)
Asian/Pacific Islander	2.0% (60)
Native Americans	0.5% (15)
Others	2.4% (70)
Unknown	6.2% (185)
Patient Location
“Central” counties of metro areas of ≥1 million population	29.5% (875)
“Fringe” counties of metro areas of ≥1 million population	24.2% (720)
Counties in metro areas of 250,000–999,999 population	20.0% (590)
Counties in metro areas of 50,000–249,999 population	7.1% (205)
Micropolitan counties	11.3% (335)
Not metropolitan or micropolitan counties	7.4% (220)
Unknown	0.5% (15)
Median Household Income
0–25th percentile	21.8% (645)
26th–50th percentile	26.9% (795)
51st–75th percentile	26.2% (775)
76th–100th percentile	22.8% (675)
Unknown	2.4% (70)
Days From Admission to Procedure	2.36 ± 2.68
Primary Payer
Medicare	71.7% (2125)
Medicaid	2.3% (65)
Primary insurance	20.6% (610)
Self-pay	0.3% (10)
No charge	0.0% (0)
Other	4.7% (140)
Unknown	0.3% (10)
Comorbidities
Alcoholism	7.8% (230)
Diabetes complicated	23.0% (680)
Drug abuse	1.4% (40)
Hypertension complicated	25.0% (740)
Chronic lung condition	16.9% (500)
Obese	26.2% (775)
Perivascular disease	8.3% (245)
Fracture Level
Cervical	31.9% (945)
Thoracic	57.2% (1695)
Lumbar	10.8% (320)
Surgical Procedure
Fusion
Cervical fusion	21.5% (635)
Cervicothoracic fusion	3.2% (95)
Thoracic fusion	44.6% (1320)
Thoracolumbar fusion	9.0% (265)
Lumbar fusion	3.7% (110)
Lumbosacral fusion	0.8% (25)
>2 levels	54.7% (1620)
Corpectomy
Cervical corpectomy	3.7% (110)
Thoracic corpectomy	10.8% (320)
Lumbar corpectomy	2.2% (65)
>2 levels	0.2% (5)
Spinal cord injury	13.3% (395)
Ankylosing Spondylitis
Cervical	13.3% (395)
Cervicothoracic	1.2% (35)
Thoracic	24.8% (735)
Thoracolumbar	2.9% (85)
Lumbar	5.2% (155)
Multiple sites	4.2% (125)
Unspecified region	19.9% (590)
Ankylosing Hyperostosis (DISH)
Cervical	7.8% (230)
Cervicothoracic	0.8% (25)
Thoracic	13.2% (390)
Thoracolumbar	1.7% (50)
Lumbar	2.0% (60)
Multiple sites	2.9% (85)
Unspecified region	6.1% (180)
Mixed (AS + DISH)	6.1% (180)
In-hospital deaths	5.4% (160)

Abbreviations: AS, ankylosing spondylitis; DISH, diffuse idiopathic skeletal hyperostosis.

View popup

Table 2

Algorithm performance in predicting in-hospital mortality.

Algorithm	Accuracy	Sensitivity	Specificity	Area Under the Curve
Adaptive Boosting Classifier	79.88%	0.400	0.8225	0.7657
Random Forest	92.74%	0.300	0.9645	0.7562
Gradient Boosting	89.94%	0.200	0.9408	0.7509
Average	87.52%	0.300	0.9093	0.7576

View popup

Table 3

Permutation feature importance for prediction of in-hospital death following surgical treatment of vertebral fractures in patients with AS and DISH.

Outcome and Features	Average Permutation Feature Importance	Mean ± SD or % (n)		P
Outcome and Features	Average Permutation Feature Importance	Survived to Discharge	In-Hospital Death	P
Days from admission to surgery	0.0851	2.36 ± 2.72	2.14 ± 1.58	0.828
Thoracic fracture	0.0843	55.2% (1635)	2.0% (60)	<0.001
Hypertension with end-organ complications	0.0541	22.5% (665)	2.5% (75)	<0.001
Spinal cord injury	0.0541	11.5% (340)	1.9% (55)	<0.001
Cervical fracture	0.0402	29.4% (870)	2.5% (75)	<0.001
Age	0.0338	72.89 ± 10.17	77.34 ± 7.69	<0.001
Primary payer - Medicare	0.0221	67.1% (1985)	4.7% (140)	<0.001
Thoracic fusion	0.0212	42.3% (1250)	2.4% (70)	0.825
Median household income: 76th–100th percentile	0.0161	22.3% (660)	0.5% (15)	<0.001
Cervical fusion	0.0156	19.8% (585)	1.69% (50)	0.002

View popup

Table 4

Gini feature importance for prediction of in-hospital death following surgical treatment of vertebral fractures in patients with AS and DISH.

Outcome and Features	Average Gini Feature Importance	Mean ± SD or % (n)		P
Outcome and Features	Average Gini Feature Importance	Survived to Discharge	In-Hospital Death	P
Age	0.172	72.89 ± 10.17	77.34 ± 7.69	<0.001
Days from admission to surgery	0.101	2.36 ± 2.72	2.14 ± 1.58	0.828
Cervical fracture	0.0646	29.4% (870)	2.5% (75)	<0.001
Spinal cord injury	0.0567	11.5% (340)	1.9% (55)	<0.001
AS of lumbar spine	0.0443	4.4% (130)	0.8% (25)	<0.001
Hypertension with end-organ complications	0.0368	22.5% (665)	2.5% (75)	<0.001
Perivascular disease	0.0363	7.6% (225)	0.7% (20)	0.046
Obesity	0.0344	25.0% (740)	1.2% (35)	0.203
Fusion >2 evels	0.0272	52.2% (1545)	2.5% (75)	0.04
Hispanic ethnicity	0.0225	7.6% (225)	0.7% (20)	0.045