Prediction of emergence from prolonged disorders of consciousness from measures within the UK rehabilitation outcomes collaborative database: a multicentre analysis using machine learning

Figures & data

Table 1. Breakdown of patients (n = 1043) on discharge, with age at admission broken down into four quartiles.

		At discharge from programme
Parameter	Overall N	Still in PDOC (FIM + FAM< =31) n= (%)	Emerged (FIM + FAM> =36) n= (%)
Gender	N = 1043	n = 579 (54%)	n = 464 (46%)
Male	689	391 (68%)	298 (64%)
Female	353	187 (32%)	166 (36%)
Time since onset (days)
Mean (95% CI)		297 (194, 468)	146 (94 232)
Aetiology diagnosis
CVA^a	309	141 (24%)	168 (36%)
Trauma	379	198 (34%)	181 (39%)
Hypoxia	241	187 (32%)	54 (12%)
Inflammatory	27	11 (2%)	16 (3%)
Tumour	18	12 (2%)	6 (1%)
Other^b	57	27 (5%)	30 (7%)
Age at admission (Years)
16–37	275	157 (27%)	118 (26%)
38–49	261	156 (27%)	105 (23%)
50–61	260	142 (26%)	118 (25%)
> =62	247	124 (22%)	132 (26%)
Mean (SD) years	1043	47.6 (15.8)	48.9 (15.9)

95% CI: 95% confidence interval.

^aCerebrovascular accidents due to infarct, haemorrhage, subarachnoid haemorrhage, or other aetiologies.

^bIncludes toxic/metabolic injuries e.g., hypoglycaemia, drug overdose, etc.

Table 2. The six variables found by LDA and their standardised canonical function coefficients.

Variables	Coefficients
NIS-Motor	0.40
NPDS-Communication	0.20
PCAT-Behaviour	−0.36
PCAT-Complex disability	0.33
PCAT-Mood	−0.24
RCS-Medical	0.31

NIS: Neurological Impairment Scale; NPDS: Northwick Park Dependency Scale; PCAT: Patient Categorisation Tool; RCS: Rehabilitation Complexity Scale.

Table 3. Independent variable importance for data-fitting ANN.

Independent variable on admission	Importance	Normalised importance
NIS-Motor	0.130	100.00%
NPDS-Communication	0.120	92.10%
NPDS-Eating	0.096	73.60%
PCAT-Behaviour	0.094	72.40%
PCAT-Psychiatric	0.094	72.40%
PCAT-Complex Disability	0.082	63.00%
RCS-Medical	0.076	58.00%
PCAT-Mood	0.067	51.50%
PCAT-Facilities	0.066	50.80%
NPDS-1:1 Specialling	0.062	47.50%
NPDS-Drinking	0.059	45.40%
PCAT-Tracheostomy	0.052	39.90%

NIS: Neurological Impairment Scale; NPDS: Northwick Park Dependency Scale; PCAT: Patient Categorisation Tool; RCS: Rehabilitation Complexity Scale.

Figure 1. ROC curve for data-fitting artificial neural networks. ROC curve for data-fitting ANN for emerged (yes) and not emerged (no) The area under the curve was 0.94 for both emerged and non-emerged on discharge.

Table 4. Rule induction—data fit six-rule set.

Rule	IF	THEN	Prediction Probability
1	NIS-Motor > 11 AND NIS-Motor ≤ 13 AND PCAT-Behaviour  ≤ 1 AND PCAT-Disability > 2	Not emerged	65%
2	NIS-Motor > 11 AND NIS-Motor ≤ 13 AND PCAT-Behaviour > 1	Emerged	75%
3	NIS-Motor ≤ 11	Emerged	75%
4	NIS-Motor > 11 AND NIS-Motor ≤ 13 AND PCAT-Behaviour ≤ 1 AND PCAT-Disability ≤ 2	Emerged	55%
5	NIS-Motor > 13 AND PCAT-Complex Disability ≤ 2	Emerged	60%
When aetiology diagnosis subcategory on admission was included:
6	NIS-Motor > 13) AND PCAT-Complex Disability > 2 AND (Diagnosis Subcategory = Anoxia OR (Stroke: Other) OR Toxic	Not emerged	91%

Data availability statement

As the UKROC data set is a live clinical data set, for reasons of confidentiality and data protection data sharing is not available at the current time. Copies of the tools used in this study are available free of charge from the authors. Please visit our website for more details and contact information. http://www.kcl.ac.uk/lsm/research/divisions/cicelysaunders/research/studies/ukroc/tools.aspx