adopting literature-based discovery on rehabilitation ...97 literature [14]. applying ddm to...

1

1 Adopting Literature-based Discovery on Rehabilitation Therapy

2 Repositioning for Stroke

3

4 Guilin Meng1,2, Yong Huang2,3, Qi Yu4, Ying Ding2, David Wild2, Yanxin Zhao1,

5 Xueyuan Liu1,* , Min Song5,*

6

7 1Tongji University School of Medicine, Shanghai Tenth People’s Hospital, Shanghai,

8 China

9 2School of Informatics Computing and Engineering, Indiana University,

10 Bloomington, IN, United States

11 3School of Information Management, Wuhan University, Wuhan, China

12 4School of management, Shanxi Medical University, Shanxi, China

13 5School of Informatics, Yonsei University, Seoul, Korea

14

15 *Corresponding author:

16 E-mail: [email protected] (LXY)

17 E-mail: [email protected] (SM)

.CC-BY 4.0 International licensewas not certified by peer review) is the author/funder. It is made available under aThe copyright holder for this preprint (whichthis version posted September 19, 2018. . https://doi.org/10.1101/422154doi: bioRxiv preprint

https://doi.org/10.1101/422154

http://creativecommons.org/licenses/by/4.0/

2

18 Abstract

19 Stroke is a common disabling disease severely affecting the daily life of the

20 patients. There is evidence that rehabilitation therapy can improve the movement

21 function. However, there are no clear guidelines that identify specific, effective

22 rehabilitation therapy schemes, and the development of new rehabilitation techniques

23 has been fairly slow. One informatics translational approach, called ABC model in

24 Literature-based Discovery, was used to mine an existing rehabilitation candidate

25 which is most likely to be repositioned for stroke. As in the classic ABC model

26 originated from Don Swanson, we built the internal links of stroke (A), assessment

27 scales (B), rehabilitation therapies (C) in PubMed relating to upper limb function

28 measurements for stroke patients. In the first step, with E-utility we retrieved both

29 stroke related assessment scales and rehabilitation therapies records, and complied

30 two datasets called Stroke_Scales and Stroke_Therapies, respectively. In the next

31 step, we crawled all rehabilitation therapies co-occurred with the Stroke_Theapies,

32 named as All_Therapies. Therapies that were already included in Stroke_Therapies

33 were deleted from All_Therapies, so that the remaining therapies were the potential

34 rehabilitation therapies, which could be repositioned for stroke after subsequent

35 filtration by manual check. We identified the top ranked repositioning rehabilitation

36 therapy following by subsequent clinical validation. Hand-arm bimanual intensive

37 training (HABIT) ranked the first in our repositioning rehabilitation therapies list,

38 with the most interaction links with Stroke_Scales. HABIT showed a significant


https://doi.org/10.1101/422154


3

39 improvement in clinical scores on assessment scales of Fugl-Meyer Assessment and

40 Action Research Arm Test in the clinical validation on upper limb function for acute

41 stroke patients. Based on the ABC model and clinical validation of the results, we put

42 forward that HABIT as a promising rehabilitation therapy for stroke, which shows

43 that the ABC model is an effective text mining approach for rehabilitation therapy

44 repositioning. The results seem to be promoted in clinical knowledge discovery.

45

46 Keywords

47 Text mining; ABC model; Stroke; Hand-arm bimanual intensive training; Upper

48 extremity

49

50 Author Summary

51 In the present study, we proposed a text mining approach to mining terms related

52 to disease, rehabilitation therapy, and assessment scale from literature, with a

53 subsequent ABC inference analysis to identify relationships of these terms across

54 publications. The clinical validation demonstrated that our approach can be used to

55 identify potential repositioning rehabilitation therapy strategies for stroke.

56 Specifically, we identified a promising rehabilitation method called HABIT

57 previously used in pediatric congenital hemiplegia. A subsequent clinical trial

58 confirmed this as a highly promising rehabilitation therapy for stroke.


https://doi.org/10.1101/422154


4

59 Introduction

60 Stroke and rehabilitation

61 Stroke is a common disabling health-care problem, which is attributed to be the

62 second-leading cause of mortality and disability worldwide [1,2]. For instance, in UK

63 stroke is the largest single cause of disability with an annual cost to society of

64 approximately £9 billion; in the United States, nearly 0.8 million people have stroke

65 annually and the estimated direct and indirect costs of stroke is $95 billion in 2015,

66 expected to rise to 185 billion in 2030 [3]. The symptoms of acute stroke include

67 physical impairments and cognitive dysfunction, and physical impairments of the

68 affected limbs range from movement restriction, sensory loss, muscle activation

69 abnormalities, etc.[4]. About 50% acute stroke survivors suffered from dysfunction of

70 the upper limbs in their chronic phase [5], severely impacting the daily life and the

71 therapeutic effect of rehabilitation therapy, which reduces the quality of life after

72 stroke [6,7].

73 Rehabilitation therapies offer a chance for an individual to recover and adapt to

74 situation following acute stroke. There has been a large amount of research into

75 methods of rehabilitation management, including task-oriented training [8], impaired

76 limb forced training [9], movement science-based therapy, robotic-assisted

77 movement, virtual reality (VR) training [10], functional electrical stimulation [11],

78 and skill acquisition training paired with impairment mitigation and motivational

79 enhancement and etc.


https://doi.org/10.1101/422154


5

80 Currently, no high-quality evidence can be found for any rehabilitation

81 management that is currently used as part of routine guideline practice, and evidence

82 is not sufficient enough to evaluate the relative effectiveness of existing rehabilitation

83 strategies in large clinical trials [12]. Furthermore, the stagnant development of new

84 competitive rehabilitation strategies impedes rehabilitation therapy repositioning from

85 possible unfolding data sources, such as large amount of literature in PubMed.

86

87 Rehabilitation therapy repositioning based on Don Swanson’s ABC model

88 Rehabilitation therapy repositioning is the application of already approved

89 therapy to new diseases，which is derived from Data-driven Method (DDM). DDM

90 is based on analyzing data about a system, in particular finding connections between

91 the system state variables (input, internal and output variables) without explicit

92 knowledge of the physical behavior of the system [13]. Scientific literature is a special

93 kind of data, usually semi-structured or unstructured, which comprises scholarly

94 publications that report original empirical and theoretical work in the natural and

95 social sciences, and within an academic field. Literature mining is a specialized data

96 mining method that is used to extract information (facts or data) from scientific

97 literature [14]. Applying DDM to literature mining can generate rehabilitation

98 therapies, which can be repositioned for stroke by systematically scrutinizing a vast

99 amount of abstracts, or full text versions of scientific publications.

100 The principal advantages of rehabilitation therapy repositioning over new therapy


https://doi.org/10.1101/422154


6

101 development are that approved therapy has already been tested for safety, and

102 repositioning can eliminate the time and cost of developing new therapy. For instance,

103 the use of VR as a rehabilitation intervention was first applied to basic motor

104 disability [15]. Research in the area of VR based rehabilitation gained growing

105 recognition of the potential value of VR for other diseases with motor disorders, for

106 example, Parkinson's disease. VR is now proposed as a new rehabilitation tool that

107 potentially optimizes motor learning in a safe environment and replicates real

108 situations to help improve functional activities in daily life such as gait, balance, and

109 quality of life [16].

110 In order to identify new repurposed rehabilitation therapies for stroke, we

111 developed a relation extraction method using ABC model proposed by Don Swanson

112 [17]. ABC model demonstrated that new knowledge could be discovered from sets of

113 disjointed scientific articles (Fig 1). In the model shown in Fig 1, one set of articles

114 (AB) reports an interesting association between variables A and B, while another set

115 of articles (BC) reports a relationship between B and C, but nothing at all has been

116 published concerning a possible link between A and C, even though such a link if

117 validated would be of scientific interests. The open ABC model is to start with theme

118 "A" in MEDLINE or PubMed that collects scientific questions (such as literatures that

119 discuss stroke); the word / phrase "B" is then listed (in the title or abstract appearing

120 in A) and a separate literature search is used for each "B" term (or filtered subset);

121 then the words and phrases "C" appearing in the code of the B are compiled (or


https://doi.org/10.1101/422154


7

122 filtered); finally, by some criteria, the C terms are ranked such that a high ranked C

123 term is said to represent the most promising hypothesis. Depending on the system, B

124 and C may represent other features extracted from medical topic titles or concepts.

125 For example, the term C may be the name of a therapy strategy that has not been

126 tested for stroke for A (stroke), but C (rehabilitation therapy) has been demonstrated

127 in other situations (e.g., in other forms of physical injury model or experimental

128 animal model) with curative effects, suggesting that C may be explored as a new

129 therapy. For each disease, there should be corresponding therapies including

130 rehabilitation therapies; meanwhile, for each rehabilitation therapy, there should be

131 corresponding measurements, most of which are assessment scales focused on

132 different aspects of rehabilitation effectiveness. So the identification of disease –

133 rehabilitation therapy, rehabilitation therapy – assessment scale and disease –

134 rehabilitation therapy relationships is key to identifying and curating new candidates

135 for rehabilitation repositioning based on the ABC model.

136

137 Fig 1. ABC model of Stroke- Assessment scales- Rehabilitation therapies

138

139 Since there are no databases that extensively curate the relationships among these

140 three entities, unidentified relationships may be buried in literature [18,19]. Recently,

141 literature mining has been applied to automatically generate various, plausible new

142 hypotheses [20,21], which motivates to find new indications of existing rehabilitation


https://doi.org/10.1101/422154


8

143 therapies for stroke.

144 In this study, with the basis of ABC model, we aimed to detect indirect

145 relationships that may facilitate the discovery of new candidates supporting the

146 curating for rehabilitation repositioning. Specifically, the aim of this study is to find

147 new rehabilitation therapy for stroke by extracting relationships from the literature

148 and to provide clinical validation to identify the most promising potential therapy.

149

150 Results

151 Text mining based on ABC model

152 We retrieved 11,418 records from PubMed with stroke related keywords, and

153 there were 10,992 records with abstract. From this dataset, 241,044 unique NPs were

154 extracted, which included 81 unique scales and 215 unique rehabilitation therapies.

155 In the potential scale list, the common phrases such as “pre- test”, “post- test”,

156 “outcome assessment” as well as other unrelated scales such as “body mass index”

157 and “depression score” and “MMSE score” were deleted after manual check from the

158 stroke related scale dataset, Stroke_Scales, which ended up 28 scales (Fig 2).

159

160 Fig 2. Stroke_Scales dataset items with frequency

161

162 FMA (fugl - meyer assessment, AS (ashworth scale), BI (barthel index), ARAT

163 (action research arm test), WMFT (wolf motor function test), RS (rankin scale), FIM


https://doi.org/10.1101/422154


9

164 (functional independence measure), MOM (main outcome measure), MI (motricity

165 index), SIAS (stroke impact scale), BBS (berg balance scale), AHA (Assisting Hand

166 Assessment), MAS (motor assessment scale), RMA (rivermead motor assessment), AI

167 (arm index), BBT (box and block test), FAT (frenchay arm test), NIHSS (National

168 Institute of Health stroke scale), POM (primary outcome measure), GAS (Goal

169 Attainment Scale), JTTHF (Jebsen-Taylor Test of Hand Function), DAS (disability

170 assessment scale), BS(brunnstrom scale), COPM (canadian occupational performance

171 measure), SMWT (Six-Minute Walk Test, VAS (visual analogue scale), PEDI

172 (Pediatric Evaluation of Disability Inventory).

173 The FMA had the largest share of 22.7%, followed by AS, BI, ARAT, WMFT,

174 RS, FIM, and MOM, whose total share was 72.8%, which means the most widely

175 used assessment scales in stroke related studies.

176 Accordingly, in the potential therapy list, the following common words were

177 deleted: “clinical practice”, “conventional therapy”, “medical therapy”, “physical

178 therapy”, “specific training”, and “combined therapy” and so on; in addition, the

179 following drug therapy of “antiplatelet therapy”, “anticoagulant therapy”,

180 “antihypertensive therapy”, and “antithrombotic therapy” were deleted. Thus, we got

181 the stroke related rehabilitation therapy dataset, Stroke_Therapies, which

182 compromising 47 rehabilitation therapies (Table 1).

183

184 Table 1. Stroke_Therapies dataset items matching with frequency


https://doi.org/10.1101/422154


10

Stroke_Therapy item Frequency Stroke_Therapy item Frequency

transcranial magnetic stimulation 495 treadmill training 15

electrical stimulation 296 peripheral nerve stimulation 14

induced movement therapy 251 bimanual training 13

robot therapy 125 gait training 13

mental practice 97 motor imagery training 13

mirror therapy 97 computer interface training 12

intensive occupational therapy 95 smart arm training 12

motor training 74 massed practice 10

somatosensory stimulation 60 transcutaneous electrical nerve stimulation 10

repetitive practice 48 vr training 10

intensive training 44 physical and occupational therapy 9

neuromuscular electrical stimulation 40 active neuromuscular stimulation 8

bilateral training 32 deep brain stimulation 7

bilateral arm training 29 functional strength training 7

noninvasive brain stimulation 27 functional task practice 7

cortical stimulation 24 motor cortex stimulation 7

tactile stimulation 24 music therapy 7

upper extremity training 24 wrist training 7

hand training 23 based mental practice training 6

neuromuscular stimulation 22 constraint induced movement therapy 6


https://doi.org/10.1101/422154


11

median nerve stimulation 20 forced use therapy 6

task practice therapy 19 paired associative stimulation 6

aerobic exercise training 17 surface neuromuscular electrical stimulation 6

unilateral training 17

185

186 With the extracted Stroke_Scales, we retrieved 60,307 records in which 60,202

187 records with abstract in PubMed. From these records, we extracted the rehabilitation

188 therapies (All_Therapies dataset), and removed those that have been applied for

189 stroke, which is listed in Table 1 (Stroke_Therapies dataset) and got the potential

190 repositioning rehabilitation therapies for stroke in Table 2. The interactions of

191 Stroke_Scales and All_Therapies dataset were shown in Fig 2.

192

193 Table 2. Potential repositioning rehabilitation therapies

Unknown rehabilitation therapies Frequency Unknown rehabilitation therapies Frequency

Cognitive behavior therapy 146 hand arm bimanual intensive training 7

massage therapy 30 intensive bimanual training 2

homeopathic treatment 16 home based intensive bimanual training 2

acupuncture treatments 12

194

195 We used these potential repositioning rehabilitation therapies with “stroke” in

196 PubMed search to exclude the records that contain the associations of those


https://doi.org/10.1101/422154


12

197 rehabilitation therapies with stroke. We found that except for “hand arm bimanual

198 intensive training” and “home based intensive bimanual training”, the rest of NPs

199 co-occur with stroke. Thus, those two previously unknown rehabilitations, “hand arm

200 bimanual intensive training” and “home based intensive bimanual training”, worth

201 further investigation. With the consideration of the practical situation where

202 rehabilitation is started in the acute phase of stroke patients, it is apparent that our

203 target population were acute stroke patients treated in hospital but not at home, so that

204 “home based intensive bimanual training” was not applied in this study, we took the

205 “hand arm bimanual intensive training” as the first choice for clinical validation.

206 In the second quadrant of Fig 3, Stroke_Scales were densely distributed and were

207 interacted with All_Therapies dataset from other three quadrants. Among them, the

208 potential repositioning rehabilitation therapies were marked in red and rose, while the

209 existing Stroke_Therapies dataset items were indicated in other colors. Among the

210 potential repositioning rehabilitation therapies, “hand arm bimanual intensive

211 training” shared the most interactions with Stroke_Scales, including Jebsen-Taylor

212 Test of Hand Function (JTTHF), Canadian Occupational, Performance Measure

213 (COPM), Assisting Hand Assessment (AHA), Pediatric Evaluation of Disability

214 Inventory (PEDI), Box and Blocks Test (BBT), Six-Minute Walk Test (SMWT), Goal

215 Attainment Scale (GAS), none of which was the commonly used stroke related

216 assessment scales according to targeting populations and disciplines difference.

217


https://doi.org/10.1101/422154


13

218 Fig 3. A. Interactions of assessment scales and all therapies; B. Interaction of

219 HABIT and assessment scales.

220

221 Hand–arm bimanual intensive training (HABIT)

222 HABIT is a bimanual intervention addressing the specific upper extremity

223 impairments in pediatric congenital hemiplegia, which is the most common physical

224 disability in childhood [22,23], typically with impairments of spasticity, sensation,

225 and reduced strength. HABIT has been reported to improve the pediatric patients

226 bimanual hand coordination and the space control of actions. Furthermore, HABIT is

227 demonstrated to be the prioritized optimal approach to improving bimanual hand use

228 and activity performance for children with hemiplegia [24], whose principles include

229 motor learning (practice specificity, types of practice, feedback) [25], and principles

230 of neuroplasticity (practice-induced brain changes arising from repetition, increasing

231 movement complexity, motivation, and reward) [26,27], which are also the key

232 contents of stroke rehabilitation functional goals.

233 We checked the frequency of different scales for stroke, and found that compared

234 with JTTHF and COPM, Fugl-Meyer assessment (FMA) was the most commonly

235 used measure for the upper limb in adult populations, almost 30% of the total

236 frequencies (Table 1), which is also in accordance with a systematic literature about

237 stroke rehabilitation studies [28]. Based on the fact that FMA was usually applied in

238 combination with action research arm test (ARAT) in clinical studies, we decided to


https://doi.org/10.1101/422154


14

239 apply FMA combined with ARAT, not COPM or JTTHF in our clinical validation.

240 The final validation of the potential candidate of repositioning rehabilitation

241 therapy was conducted in the clinical trial of stroke patients with upper limb

242 impairment. The trial was registered with ChiClinicalTrials.gov, number

243 CTR-INR-1701046 [29]. To our best knowledge, there have been no other clinical

244 studies besides ours evaluating HABIT as a therapy for adult acute stroke patients.In

245 general, patients were assessed by FMA and ARAT for motor function and extremity

246 activity pre- and post-2 consecutive weeks of HABIT therapy. As shown in Table 3,

247 after the rehabilitation therapy, a direct comparison of FMA and ARAT scores

248 revealed the significant improvement, showing that HABIT improved the scores in

249 both scales.

250

251 Table 3. FMA and ARAT scores pre- and post- HABIT rehabilitation therapy

Scales Pre-therapy Post-therapy P

FMA 33.25±5.89 51.73±6.44 <0.001

ARAT 30.31±6.07 34.47±6.22 <0.001

252

253 Discussion

254 ABC model in rehabilitation therapy repositioning

255 Knowledge is sometimes segregated by syntactically impenetrable keyword

256 barriers or undiscovered in an entirely different research corpus, so that clinicians in


https://doi.org/10.1101/422154


15

257 Neurology domain may not be able to keep completely up-to-date with applicable

258 rehabilitation therapies for stroke patients. Analyzing the literature and data to

259 semi-automatically generate a hypothesis about rehabilitation therapy repositioning

260 might become the de facto approach to inform clinicians who are trying to master the

261 exponentially rapid expansion of publications and datasets. Swanson [17] proposed

262 the ABC model that can be applied to new hypothesis generation for rehabilitation

263 therapy repositioning , where ABC model is pertinent to an association rule between a

264 separate set of publications: if A is associated with B, and B is associated with C, then

265 there is a potential relation between A and C. ABC model has played a number of

266 roles in the direction of drug discovery and repositioning. Earliest applications of the

267 ABC model derived from two major findings of fish oil treatment for Raynaud's

268 disease and magnesium treatment for migraines, both of which have been clinically

269 confirmed [30]; in recent years, vigorous development of bioinformatics mining and

270 omics study indirectly borrowed the mode l [31]. However, one biggest limitation of

271 the current ABC model-based approaches is that without certain domain knowledge, it

272 is not easy to identify significative AB and BC directly. In our study, the neurology

273 clinicians (domain experts) were fully engaged, including raising questions, checking

274 manually and confirming the final candidate on the basis of expertise， which

275 eliminated the limitation

276 On the other hand, although the repositioning of rehabilitation therapy is not

277 uncommon in clinical practice, this discovery is often based on the clinician's personal


https://doi.org/10.1101/422154


16

278 experience or deduction from the medical community, without objective, systemic

279 approach to data mining application. In our discovery case, stroke (A) is associated

280 with the assessment scales (B) in stroke literature, and in the rehabilitation therapy

281 literature, assessment scales (B) represent the effect of rehabilitation therapy (C),

282 then, it is highly likely that the retrieved rehabilitation therapies that are unknown for

283 stroke yet have a positive effect on stroke. It is the first application of ABC model to

284 show the repositioning of rehabilitation therapy with positive validation. This model

285 could be generalized as disease-assessment scale-rehabilitation therapy in future

286 studies.

287 HABIT helps upper extremity function recovery for stroke

288 In the clinical validation, the patient group who received HABIT demonstrated

289 significant improvement indicating that HABIT has a positive impact on rehabilitation

290 therapy for upper extremity impaired patients.

291 The principle of HABIT includes motor learning (task specificity, task type and

292 feedback) and neuroplasticity, as well as brain transformation upon increasingly

293 difficult therapy and incentive reward [26], training cortico-spinal system

294 reconstruction, manifesting as function recovery after injury [32], which is the

295 theoretical foundation of hypothesizing that bimanual intensive therapy could

296 improve the upper extremity function after acute stroke.

297 In the present study, we proposed a text mining approach to mining terms related

298 to disease, rehabilitation therapy, and assessment scale from literature, with a


https://doi.org/10.1101/422154


17

299 subsequent ABC inference analysis to identify relationships of these terms across

300 publications. The clinical validation demonstrated that our approach can be used to

301 identify potential repositioning rehabilitation therapy strategies for stroke.

302 Specifically, we identified a promising rehabilitation method called HABIT

303 previously used in pediatric congenital hemiplegia. A subsequent clinical trial

304 confirmed this as a highly promising rehabilitation therapy for stroke. As a follow-up

305 study, moreclinical trials should be involved for the long-term impact of HABIT on

306 stroke patients, and optimal parameterization of the therapy. We also plan to refine

307 the text mining and inference strategy and apply to other clinical applications

308 amenable to the technique.

309

310 Materials and methods

311 Study procedure

312 In ABC model, there is an internal connection among disease, rehabilitation

313 therapy, and assessment scale. For stroke, most assessment scales of upper limb

314 impairment rehabilitation assess a person's ability to manage daily activities that

315 require the use of the upper limbs, whatever the therapy strategies involved;

316 meanwhile, those rehabilitation therapies for functional improvement of upper limb

317 share the same set of scales for assessment, regardless of whether they are applicable

318 of stroke or not yet. So our plan was to identify undiscovered rehabilitation therapies

319 for stroke through shared assessment scales (B), and ultimately clinically validate the


https://doi.org/10.1101/422154


18

320 most promising candidate.

321

322 ABC model implementation

323 Our first focus was to find repositioning rehabilitation therapy candidates from an

324 extensive collection of articles in PubMed. The flow chart was shown in Fig 4.

325 (The code of this article can be found in

326 https://github.com/hyyc116/Stroke_findings/tree/master/DSTN.)

327

328 Fig 4. Flow chart from stroke to repurposed rehabilitation therapy

329

330 Stroke related assessment scales and rehabilitation therapy datasets creation. To

331 collect articles related to stroke with upper limb impairment, we searched PubMed

332 with stroke related keywords: (“stroke” OR “cerebral infarction” OR “brain ischemia”

333 OR “cerebral hemorrhagic” OR “subarachnoid hemorrhage”) and (“hand” OR “arm”

334 OR “upper extremity” OR “upper limb”), with the proviso of humans.

335 The reason why we did not use Medical Subject Headings (MeSH) terms is that

336 most specific assessment scales and therapies do not directly belong to MeSH terms.

337 When we searched stroke related rehabilitation items, the MeSH terms we got were

338 only the common words (see complementary material “MeSH terms”), which should

339 be deleted as confounding factors.

340 E-utilities (https://www.ncbi.nlm.nih.gov/books/NBK25500/) were used to fetch


https://doi.org/10.1101/422154


19

341 all query related data in the PubMed. Scales and rehabilitation therapies are always

342 noun phrases (NPs) in scientific articles. Thus, we applied a shallow chunk analyzer

343 to extract NPs. The NPs ended with “test”, “scale”, “assessment”, “measure”, “score”

344 or “index” with frequency more than 5 were kept to be a possible stroke assessment

345 scale candidate. In addition, NPs ended with “training”, “therapy”, “treatment”,

346 “treatments”, “practice”, “program”, “practise” or “simulation” with frequency more

347 than 5 as well were kept to be a possible stroke therapy candidate.

348 Since relying only on the predefined lexicon to gather rehabilitation documents

349 could lead to false negatives, manual inspection should be conducted to identify true

350 positives as accurately as possible before stroke related scale dataset (Stroke_Scales)

351 and rehabilitation therapy dataset (Stroke_Therapies) were established.

352

353 Entire rehabilitation therapy dataset creation. ABC model has been successful in

354 explaining how two concepts are linked by an intermediate therapy discovery [17].

355 Specifically, with a direct stroke - assessment scales and therapies - assessment scales

356 relationship, we crawled all therapies related NP in PubMed co-occurred with

357 Stroke_Scales via E-Utility. The proceeding entire rehabilitation therapy dataset,

358 named All_Therapies building was similar with Stroke_Scales building mentioned

359 above, that the NPs ended with “training”, “therapy”, “treatment”, “treatments”,

360 “practice”, “program”, “practice” or “simulation” with frequency more than 5 were

361 kept to be a possible therapy candidate in which the assessment was the item in


https://doi.org/10.1101/422154


20

362 Stroke_Scales. Manual inspection was the same as above of Stroke_Therapies.

363

364 Potential repositioning rehabilitation therapies dataset. Therapies already included

365 in Stroke_Therapies were deleted from All_Therapies, so that the remaining therapies

366 were not stroke-applied therapies, which could be repurposed for stroke. Manual

367 inspection was carried out again.

368

369 Hypotheses validation

370 Validation of the retrieved Repositioning_Therapies dataset in PubMed. The

371 articles related to potential repositioning rehabilitation therapies together with stroke

372 related keywords were retrieved from PubMed to ensure that there are no articles that

373 contain associations of therapies with stroke.

374

375 Further rehabilitation theory of potential candidate exploration. The knowledge

376 discovery is full of uncertainty and complicated, in which algorithms and methods

377 could be perfect in theory, while the precision, recall or some other metrics could be

378 meaningless to some extends. Thus, what we aim at is to find potential candidates.

379 Clinical value of the potential candidates then should be further comprehensively

380 explored in the mechanism and principles with rehabilitation theory.

381

382 Validation in clinical trials. A clinical trial of adult acute stroke patients was carried


https://doi.org/10.1101/422154


21

383 out to test the rehabilitation effect by analyzing several sorts of data, aiming to clarify

384 the potential advantage of rehabilitation therapy and to determine the optimal

385 rehabilitation approach for stroke patients.

386

387 Acknowledgments

388 This study was supported by the National Natural Science Foundation of China

389 (81771133, 71573162) and also partly supported by the Bio-Synergy Research Project

390 (NRF-2013M3A9C4078138) of the Ministry of Science, ICT and Future Planning

391 through the National Research Foundation.

392


https://doi.org/10.1101/422154


22

393 References

394 1. Cramer SC, Wolf SL, Adams HP, Jr., Chen D, Dromerick AW, Dunning K, et al.

395 Stroke Recovery and Rehabilitation Research: Issues, Opportunities, and the National

396 Institutes of Health StrokeNet. Stroke. 2017; 48: 813-819.

397 2. Brainin M, Zorowitz RD. Advances in stroke: recovery and rehabilitation. Stroke.

398 2013; 44: 311-313.

399 3. The L. Stroke-acting FAST at all ages. Lancet. 2018; 391: 514.

400 4. Kuehn B. Stroke Rehab Lacking. JAMA. 2018; 320: 128.

401 5. Favre I, Zeffiro TA, Detante O, Krainik A, Hommel M, Jaillard A. Upper limb

402 recovery after stroke is associated with ipsilesional primary motor cortical activity: a

403 meta-analysis. Stroke. 2014; 45: 1077-1083.

404 6. Han CE, Arbib MA, Schweighofer N. Stroke rehabilitation reaches a threshold.

405 PLoS Comput Biol. 2008; 4: e1000133.

406 7. Franck JA, Smeets R, Seelen HAM. Changes in arm-hand function and arm-hand

407 skill performance in patients after stroke during and after rehabilitation. PLoS One.

408 2017; 12: e0179453.

409 8. Carrico C, Chelette KC, 2nd, Westgate PM, Powell E, Nichols L, Fleischer A, et al.

410 Nerve Stimulation Enhances Task-Oriented Training in Chronic, Severe Motor

411 Deficit After Stroke: A Randomized Trial. Stroke. 2016; 47: 1879-1884.

412 9. Kwakkel G, Veerbeek JM, van Wegen EE, Wolf SL. Constraint-induced movement

413 therapy after stroke. Lancet Neurol. 2015; 14: 224-234.


https://doi.org/10.1101/422154


23

414 10. Brunner I, Skouen JS, Hofstad H, Assmus J, Becker F, Sanders AM, et al. Virtual

415 Reality Training for Upper Extremity in Subacute Stroke (VIRTUES): A multicenter

416 RCT. Neurology. 2017; 89: 2413-2421.

417 11. Kattenstroth JC, Kalisch T, Sczesny-Kaiser M, Greulich W, Tegenthoff M, Dinse

418 HR. Daily repetitive sensory stimulation of the paretic hand for the treatment of

419 sensorimotor deficits in patients with subacute stroke: RESET, a randomized,

420 sham-controlled trial. BMC Neurol. 2018; 18: 2.

421 12. Pollock A, Farmer SE, Brady MC, Langhorne P, Mead GE, Mehrholz J, et al.

422 Interventions for improving upper limb function after stroke. Cochrane Database Syst

423 Rev. 2014: CD010820.

424 13. Intosalmi J, Nousiainen K, Ahlfors H, Lahdesmaki H. Data-driven mechanistic

425 analysis method to reveal dynamically evolving regulatory networks. Bioinformatics.

426 2016; 32: i288-i296.

427 14. Chung D, Lawson A, Zheng WJ. A statistical framework for biomedical literature

428 mining. Stat Med. 2017; 36: 3461-3474.

429 15. Greenleaf WJ, Tovar MA. Augmenting reality in rehabilitation medicine. Artif

430 Intell Med. 1994; 6: 289-299.

431 16. Dockx K, Bekkers EM, Van den Bergh V, Ginis P, Rochester L, Hausdorff JM, et

432 al. Virtual reality for rehabilitation in Parkinson's disease. Cochrane Database Syst

433 Rev. 2016; 12: CD010760.

434 17. Swanson DR, Smalheiser NR. An interactive system for finding complementary


https://doi.org/10.1101/422154


24

435 literatures: A stimulus to scientific discovery. Artificial Intelligence. 1997; 91:

436 183-203.

437 18. Ding Y, Liu XZ, Guo C, Cronin B. The distribution of references across texts:

438 Some implications for citation analysis. Journal of Informetrics. 2013; 7: 583-592.

439 19. Westergaard D, Staerfeldt HH, Tonsberg C, Jensen LJ, Brunak S. A

440 comprehensive and quantitative comparison of text-mining in 15 million full-text

441 articles versus their corresponding abstracts. PLoS Comput Biol. 2018; 14: e1005962.

442 20. He B, Tang J, Ding Y, Wang H, Sun Y, Shin JH, et al. Mining relational paths in

443 integrated biomedical data. PLoS One. 2011; 6: e27506.

444 21. Percha B, Altman RB. Learning the Structure of Biomedical Relationships from

445 Unstructured Text. PLoS Comput Biol. 2015; 11: e1004216.

446 22. Gordon AM, Schneider JA, Chinnan A, Charles JR. Efficacy of a hand-arm

447 bimanual intensive therapy (HABIT) in children with hemiplegic cerebral palsy: a

448 randomized control trial. Dev Med Child Neurol. 2007; 49: 830-838.

449 23. Charles J, Gordon AM. Development of hand-arm bimanual intensive training

450 (HABIT) for improving bimanual coordination in children with hemiplegic cerebral

451 palsy. Dev Med Child Neurol. 2006; 48: 931-936.

452 24. Green D, Schertz M, Gordon AM, Moore A, Schejter Margalit T, Farquharson Y,

453 et al. A multi-site study of functional outcomes following a themed approach to

454 hand-arm bimanual intensive therapy for children with hemiplegia. Dev Med Child

455 Neurol. 2013; 55: 527-533.


https://doi.org/10.1101/422154


25

456 25. Schmidt RA, Lee TD. Motor control and learning: A behavioral emphasis (4th

457 ed.). 2005:

458 26. Nudo RJ. Adaptive plasticity in motor cortex: implications for rehabilitation after

459 brain injury. J Rehabil Med. 2003: 7-10.

460 27. Kleim JA, Hogg TM, VandenBerg PM, Cooper NR, Bruneau R, Remple M.

461 Cortical synaptogenesis and motor map reorganization occur during late, but not

462 early, phase of motor skill learning. J Neurosci. 2004; 24: 628-633.

463 28. Santisteban L, Teremetz M, Bleton JP, Baron JC, Maier MA, Lindberg PG. Upper

464 Limb Outcome Measures Used in Stroke Rehabilitation Studies: A Systematic

465 Literature Review. PLoS One. 2016; 11: e0154792.

466 29. Meng G, Meng X, Tan Y, Yu J, Jin A, Zhao Y, et al. Short-term Efficacy of

467 Hand-Arm Bimanual Intensive Training on Upper Arm Function in Acute Stroke

468 Patients: A Randomized Controlled Trial. Front Neurol. 2017; 8: 726.

469 30. Swanson DR. Fish oil, Raynaud's syndrome, and undiscovered public knowledge.

470 Perspect Biol Med. 1986; 30: 7-18.

471 31. Smalheiser NR. Rediscovering Don Swanson: the Past, Present and Future of

472 Literature-Based Discovery. J Data Inf Sci. 2017; 2: 43-64.

473 32. Eyre JA. Development and plasticity of the corticospinal system in man. Neural

474 Plast. 2003; 10: 93-106.

475


https://doi.org/10.1101/422154



https://doi.org/10.1101/422154


adopting literature-based discovery on rehabilitation ...97 literature [14]. applying ddm to...

Documents