Table 2 Selected variables to predict mortality risk based on the training set (N = 18,682) using Elastic-net logistic regression (“glmnet” R package), SCAD- support vector machine (“penalizedSVM” R package), and MCP- neural network algorithm with repeated tenfold cross-validation in the training set (“neuralnet” and “ncvreg” R packages).

From: Machine learning algorithms for identifying predictive variables of mortality risk following dementia diagnosis: a longitudinal cohort study

Algorithm	Selected variables	Normalized importance value %	P-value
Elastic net-regularized logistic regression (Alpha hyperparameter:0.5, lambda tuning’s parameter:0.006196)	Age at dementia diagnosis	100	< 0.001
	MMSE score	85	< 0.001
	BMI	74	< 0.001
	Basic dementia diagnostic work-up	56	< 0.001
	Diuretics	45	< 0.001
	Time from referral to initiation of work-up	42	0.367
	Time from initiation of work-up to diagnosis	37	< 0.001
	Atorvastatin	35	< 0.001
	Statins	35	< 0.001
	Cholinesterase inhibitors	32	< 0.001
	Rosuvastatin	32	< 0.001
	Heart failure	31	0.012
	Charlson comorbidity index	27	< 0.001
	Dementia medications	25	< 0.001
	Physiotherapist assessment	22	< 0.001
	Place of residency	21	0.001
	Fluvastatin	21	0.421
	Alcohol related diagnosis	17	< 0.001
	Renal disease	15	< 0.001
	Total number of medications at baseline	14	< 0.001
	Municipality	13	0.626
	Sex	12	< 0.001
	Liver failure	12	0.062
	Atrial fibrillation	11	< 0.001
	Anemia	11	< 0.001
	Ischemic heart failure	10	< 0.001
	Dementia type	9	< 0.001
	Diabetes mellitus	8	< 0.001
	Renin-angiotensin system inhibitors	8	0.635
	Renin-angiotensin system inhibitors two or more years before dementia diagnosis	8	0.449
	Losartan	6	0.001
	Antidepressants	3	0.135
	Irbesartan	3	0.78
	Acute Kidney Injury	2	0.003
	Cancer	2	0.002
	Stroke	2	< 0.001
	Captopril	2	0.038
	Calcium channel blockers	2	0.367
SCAD-regularized SVM (Kernel: sigmoid, Number of Support Vectors: 18,673, lambda tuning’s parameter:0.010)	Age at dementia diagnosis	100	< 0.001
	MMSE score	90	< 0.001
	Basic dementia diagnostic work-up	85	< 0.001
	BMI	83	< 0.001
	Time from referral to diagnosis	68	< 0.001
	Time from referral to initiation of work-up	67	0.367
	Time from initiation of work-up to diagnosis	64	< 0.001
	Physiotherapist assessment	50	< 0.001
	Charlson comorbidity index	39	< 0.001
	Diuretics	35	< 0.001
	Sex	33	< 0.001
	Dementia medications	25	< 0.001
	Atorvastatin	25	< 0.001
	Heart failure	23	0.012
	Care unit (primary care vs specialist care)	22	< 0.001
	Rosuvastatin	21	< 0.001
	Statins	18	< 0.001
	Simvastatin	17	0.722
	Total number of medications at baseline	14	< 0.001
	Diabetes mellitus	7	< 0.001
	Renin-angiotensin system inhibitors two or more years before dementia diagnosis	7	0.449
	Hypertension	5	< 0.001
	Acute Kidney Injury	3	0.003
	Cancer	2	0.002
	Liver failure	2	0.062
MCP-regularized Backpropagation Neural Network (Softmax activation function, lambda tuning’s parameter:0.0020)	Age at dementia diagnosis	100	< 0.001
	BMI	89	< 0.001
	MMSE score	75	< 0.001
	Diuretics	51	< 0.001
	Time from referral to initiation of work-up	44	0.367
	Time from initiation of work-up to diagnosis	42	< 0.001
	Atorvastatin	34	< 0.001
	Basic dementia diagnostic work-up	30	< 0.001
	Sex	25	< 0.001
	Charlson comorbidity index	21	< 0.001
	Dementia medications	20	< 0.001
	Municipality	18	0.626
	Total number of medications at baseline	16	< 0.001
	Heart failure	16	0.012
	Physiotherapist assessment	14	< 0.001
	Care unit (primary care vs specialist care)	14	< 0.001
	Cholinesterase inhibitors	13	< 0.001
	Atrial fibrillation	13	< 0.001
	Rosuvastatin	12	< 0.001
	Place of residency	11	0.001
	Alcohol related diagnosis	10	< 0.001
	Renin-angiotensin system inhibitors	9	0.635
	Renin-angiotensin system inhibitors two or more years before dementia diagnosis	9	0.449
	Dementia type	8	< 0.001
	Diabetes mellitus	8	< 0.001
	Blood tests	7	0.902
	Losartan	7	0.001
	Statins	7	< 0.001
	Antidepressants	4	0.135
	Cardiovascular medication at diagnosis	4	< 0.001
	Renal disease	4	< 0.001
	Cancer	3	0.002
	Irbesartan	3	0.78
	Liver failure	3	0.062
	Captopril	2	0.038
	Valsartan	2	0.096
	Calcium channel blockers	2	0.367
	Heart failure	2	0.359
	Fluvastatin	1	0.421
	Anemia	1	< 0.001

P-values were reported using simple logistic regression, importance values were reported based on Gini index.

Back to article page

Search

Quick links