Improving Medication Regimen Recommendation for Parkinson’s Disease Using Sensor Technology
<p>Sample PKG sensor time-series data [<a href="#B23-sensors-21-03553" class="html-bibr">23</a>], displaying an individual’s change in dyskinesia and bradykinesia scores in response to medication with the median, 25th, and 75th percentile, compared to a non-PD control group averaged over six days. (<b>A</b>) The dyskinesia time-series, (<b>B</b>) the bradykinesia time-series, and (<b>C</b>) the patient’s self-reported acknowledgment of medication administration.</p> "> Figure 2
<p>The study design is divided into two parts as labeled by the blocks “Clustering” and “Prediction”. In the clustering block, (<b>A</b>) we begin with a diverse cohort of PD patients; (<b>B</b>) each patient is assessed by MDS-UPDRS-III scores and PKG summary dyskinesia and bradykinesia scores; the physician icon is greyed out as in the future for some remote contexts this may be accomplished using only the PKG time-series data, but we currently collect both for validation purposes; (<b>C</b>) we identify similar medication regimen clusters through k-means clustering. These clusters are used in the prediction block; (<b>D</b>) we optimize medication regimens and perform statistical analysis on demographic similarities for each group—to create a decision support tool to provide enhanced initial regimen estimates; (<b>E</b>) machine learning methods, specifically random forest, are applied to predict an unknown patient’s optimized regimen cluster based on physician assessment and/or wearable sensor measurements depending on the context; (<b>F</b>) the new patient’s data will be incorporated to improve the accuracy of the decision support tool.</p> "> Figure 3
<p>(<b>a</b>,<b>b</b>) demonstrates the clusters for the motor function changes between visit 1 to visit 2 based on the MDS-UPDRS-III scores and the PKG’s summary dyskinesia and bradykinesia scores, respectively. (<b>c</b>,<b>d</b>) highlight the subjects’ best symptom control recorded using MDS-UPDRS-III scores and PKG’s summary dyskinesia and bradykinesia scores, respectively. The large shapes denote each cluster’s centroid, and the exterior marker of each point corresponds to the cluster centroid shape. The capital letters (A–D) are used to refer to each cluster. Each point’s interior maker in (<b>a</b>,<b>b</b>) represents the state of MDS-UPDRS-III and PKG change for each patient, where a black interior-point denotes the patient stayed the same between visits, a green point denotes that the patient’s visit 2 scores were better than their visit 1 scores. A red point denotes that the patient’s visit 2 scores were worse than their visit 1 scores. Note that due to the three-dimensional projection of the plot, the distance between points may appear skewed. See Supplemental Information for a 3-D animation showcasing the clusters’ position in space.</p> "> Figure A1
<p>WCSS as a function of the number of clusters when clustering patients based on their best symptom control medication regimens between the two visits according to PKG’s summary dyskinesia and bradykinesia scores. The number of clusters is set to four, as a number towards the bottom of the “elbow” is considered best, but further increasing the number of clusters has been shown to reduce the model’s robustness.</p> ">
Abstract
:1. Introduction
2. Materials and Methods
2.1. Study Cohort
2.2. Study Design
2.3. K-means Clustering
2.4. Random Forest
3. Results
3.1. Cohort Characteristics
3.2. Patient Clustering Using Medication Regimen
3.3. Random Forest Classification Using PKG Readouts
4. Discussion
Limitations
5. Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Conflicts of Interest
Appendix A
Calculated Features | Description |
---|---|
Dyskinesia | |
Dyskinesia ar coefficient k 10 coeff 2 | Unconditional maximum likelihood of an autoregressive AR(k) process—coeff 2—k10 |
Dyskinesia fft coefficient coeff 39 attr “angle” | One-dimensional discrete fast Fourier transform—coeff 39—angle |
Dyskinesia spkt welch density coeff 5 | Cross power spectral density—coeff 5 |
Bradykinesia | |
Bradykinesia agg autocorrelation f agg “mean” maxlag 40 | Aggregation function fagg—mean maxlag—40 |
Bradykinesia agg linear trend f agg “mean” chunk len 50 att “rvalue” | Linear least-squares regression aggregated over chunks versus the sequence from 0 up to the number of chunks minus one—mean—chunk length 50—attribute rvalue |
bradykinesia agg linear trend f agg “min” chunk len 50 attr “slope” | Linear least-squares regression aggregated over chunks versus the sequence from 0 up to the number of chunks minus one—mean—chunk length 50—attribute slope |
Bradykinesia fft coefficient coeff 23 attr “abs” | One-dimensional discrete fast Fourier transform—coeff 23—abs |
Bradykinesia fft coefficient coeff 94 attr “angle” | One-dimensional discrete fast Fourier transform—coeff 94—angle |
Bradykinesia fft coefficient coeff 76 attr “imag” | One-dimensional discrete fast Fourier transform—coeff 76—imag |
Bradykinesia index mass quantile q 0.9 | Relative index i where q% of the mass of the time series x lie left of I-quantile 0.9 |
References
- Schapira, A.H.; Chaudhuri, K.R.; Jenner, P. Non-motor features of Parkinson disease. Nat. Rev. Neurosci. 2017, 18, 435. [Google Scholar] [CrossRef]
- Feigin, V.L.; Abajobir, A.A.; Abate, K.H.; Abd-Allah, F.; Abdulle, A.M.; Abera, S.F.; Abyu, G.Y.; Ahmed, M.B.; Aichour, A.N.; Aichour, I. Global, regional, and national burden of neurological disorders during 1990–2015: A systematic analysis for the Global Burden of Disease Study 2015. Lancet Neurol. 2017, 16, 877–897. [Google Scholar] [CrossRef] [Green Version]
- Dorsey, E.R.; George, B.P.; Leff, B.; Willis, A.W. The coming crisis: Obtaining care for the growing burden of neurodegenerative conditions. Neurology 2013, 80, 1989–1996. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Kurtzke, J.F.; Goldberg, I.D. Parkinsonism death rates by race, sex, and geography. Neurology 1988, 38, 1558. [Google Scholar] [CrossRef]
- Lanska, D.J. The geographic distribution of Parkinson’s disease mortality in the United States. J. Neurol. Sci. 1997, 150, 63–70. [Google Scholar] [CrossRef]
- Mayeux, R.; Marder, K.; Cote, L.J.; Denaro, J.; Hemenegildo, N.; Mejia, H.; Tang, M.X.; Lantigua, R.; Wilder, D.; Gurland, B. The frequency of idiopathic Parkinson’s disease by age, ethnic group, and sex in northern Manhattan, 1988–1993. Am. J. Epidemiol. 1995, 142, 820–827. [Google Scholar] [CrossRef] [PubMed]
- Dorsey, E.R.; Venkataraman, V.; Grana, M.J.; Bull, M.T.; George, B.P.; Boyd, C.M.; Beck, C.A.; Rajan, B.; Seidmann, A.; Biglan, K.M. Randomized controlled clinical trial of “virtual house calls” for Parkinson disease. JAMA Neurol. 2013, 70, 565–570. [Google Scholar] [CrossRef] [Green Version]
- Willis, A.; Schootman, M.; Evanoff, B.; Perlmutter, J.; Racette, B. Neurologist care in Parkinson disease: A utilization, outcomes, and survival study. Neurology 2011, 77, 851–857. [Google Scholar] [CrossRef] [Green Version]
- Dahodwala, N.; Karlawish, J.; Siderowf, A.; Duda, J.E.; Mandell, D.S. Delayed Parkinson’s disease diagnosis among African-Americans: The role of reporting of disability. Neuroepidemiology 2011, 36, 150–154. [Google Scholar] [CrossRef] [Green Version]
- Saunders-Pullman, R.; Wang, C.; Stanley, K.; Bressman, S.B. Diagnosis and referral delay in women with Parkinson’s disease. Gend. Med. 2011, 8, 209–217. [Google Scholar] [CrossRef] [Green Version]
- Olanow, C.W.; Obeso, J.A.; Stocchi, F. Continuous dopamine-receptor treatment of Parkinson’s disease: Scientific rationale and clinical implications. Lancet Neurol. 2006, 5, 677–687. [Google Scholar] [CrossRef]
- Goetz, C.G.; Tilley, B.C.; Shaftman, S.R.; Stebbins, G.T.; Fahn, S.; Martinez-Martin, P.; Poewe, W.; Sampaio, C.; Stern, M.B.; Dodel, R. Movement Disorder Society-sponsored revision of the Unified Parkinson’s Disease Rating Scale MDS-UPDRS: Scale presentation and clinimetric testing results. Mov. Disord. Off. J. Mov. Disord. Soc. 2008, 23, 2129–2170. [Google Scholar] [CrossRef]
- Hauser, R.A.; Russ, H.; Haeger, D.A.; Bruguiere-Fontenille, M.; Müller, T.; Wenning, G.K. Patient evaluation of a home diary to assess duration and severity of dyskinesia in Parkinson disease. Clin. Neuropharmacol. 2006, 29, 322–330. [Google Scholar] [CrossRef]
- Ramdhani, R.A.; Khojandi, A.; Shylo, O.; Kopell, B.H. Optimizing clinical assessments in Parkinson’s disease through the use of wearable sensors and data driven modeling. Front. Comput. Neurosci. 2018, 12, 72. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Maetzler, W.; Klucken, J.; Horne, M. A clinical view on the development of technology-based tools in managing Parkinson’s disease. Mov. Disord. 2016, 31, 1263–1271. [Google Scholar] [CrossRef] [PubMed]
- Keijsers, N.L.; Horstink, M.W.; Gielen, S.C. Ambulatory motor assessment in Parkinson’s disease. Mov. Disord. Off. J. Mov. Disord. Soc. 2006, 21, 34–44. [Google Scholar] [CrossRef] [Green Version]
- Tsipouras, M.G.; Tzallas, A.T.; Rigas, G.; Bougia, P.; Fotiadis, D.I.; Konitsiotis, S. Automated levodopa-induced dyskinesia assessment. In Proceedings of the 2010 Annual International Conference of the IEEE Engineering in Medicine and Biology, Buenos Aires, Argentina, 31 August–4 September 2010; pp. 2411–2414. [Google Scholar]
- Griffiths, R.I.; Kotschet, K.; Arfon, S.; Xu, Z.M.; Johnson, W.; Drago, J.; Evans, A.; Kempster, P.; Raghav, S.; Horne, M.K. Automated assessment of bradykinesia and dyskinesia in Parkinson’s disease. J. Parkinsons Dis. 2012, 2, 47–55. [Google Scholar] [CrossRef] [PubMed] [Green Version]
- Pulliam, C.L.; Heldman, D.A.; Brokaw, E.B.; Mera, T.O.; Mari, Z.K.; Burack, M.A. Continuous assessment of levodopa response in Parkinson’s disease using wearable motion sensors. IEEE Trans. Biomed. Eng. 2017, 65, 159–164. [Google Scholar] [CrossRef] [PubMed]
- Kotschet, K.; Johnson, W.; McGregor, S.; Kettlewell, J.; Kyoong, A.; O’Driscoll, D.M.; Turton, A.R.; Griffiths, R.; Horne, M.K. Daytime sleep in Parkinson’s disease measured by episodes of immobility. Parkinsonism Relat. Disord. 2014, 20, 578–583. [Google Scholar] [CrossRef]
- Horne, M.K.; McGregor, S.; Bergquist, F. An objective fluctuation score for Parkinson’s disease. PLoS ONE 2015, 10. [Google Scholar] [CrossRef]
- Braybrook, M.; O’Connor, S.; Churchward, P.; Perera, T.; Farzanehfar, P.; Horne, M. An ambulatory tremor score for Parkinson’s disease. J. Parkinsons Dis. 2016, 6, 723–731. [Google Scholar] [CrossRef] [Green Version]
- Nahab, F.B.; Abu-Hussain, H.; Moreno, L. Evaluation of clinical utility of the Personal KinetiGraph® in the management of Parkinson’s disease. Adv. Parkinsons Dis. 2019, 8, 42. [Google Scholar] [CrossRef] [Green Version]
- Marras, C.; Chaudhuri, K.R.; Titova, N.; Mestre, T.A. Therapy of Parkinson’s disease subtypes. Neurotherapeutics 2020, 17, 1366–1377. [Google Scholar] [CrossRef]
- Charvin, D.; Medori, R.; Hauser, R.A.; Rascol, O. Therapeutic strategies for Parkinson disease: Beyond dopaminergic drugs. Nat. Rev. Drug Discov. 2018, 17, 804–822. [Google Scholar] [CrossRef] [PubMed]
- Van Rooden, S.M.; Heiser, W.J.; Kok, J.N.; Verbaan, D.; Van Hilten, J.J.; Marinus, J. The identification of Parkinson’s disease subtypes using cluster analysis: A systematic review. Mov. Disord. 2010, 25, 969–978. [Google Scholar] [CrossRef]
- Yock, A.D.; Kim, G.Y. Using k-means clustering to determine the number and position of isocenters in MLC-based multiple target intracranial radiosurgery. J. Appl. Clin. Med. Phys. 2017, 18, 351–357. [Google Scholar] [CrossRef]
- Breiman, L. Random forests. Mach. Learn. 2001, 45, 5–32. [Google Scholar] [CrossRef] [Green Version]
- Liaw, A.; Wiener, M. Classification and regression by randomForest. R News 2002, 2, 18–22. [Google Scholar]
- Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.; Dubourg, V. Scikit-learn: Machine learning in python. J. Mach. Learn. Res. 2011, 12, 2825–2830. [Google Scholar]
- Tan, P.; Steinbach, M.; Kumar, V. Introduction to Data Mining; Pearson Education: Boston, MA, USA, 2006. [Google Scholar]
- Pahwa, R.; Isaacson, S.H.; Torres-Russotto, D.; Nahab, F.B.; Lynch, P.M.; Kotschet, K.E. Role of the Personal KinetiGraph in the routine clinical assessment of Parkinson’s disease: Recommendations from an expert panel. Expert Rev. Neurother. 2018, 18, 669–680. [Google Scholar] [CrossRef]
- Mu, J.; Chaudhuri, K.R.; Bielza, C.; de Pedro-Cuesta, J.; Larrañaga, P.; Martinez-Martin, P. Parkinson’s disease subtypes identified from cluster analysis of motor and non-motor symptoms. Front. Aging Neurosci. 2017, 9, 301. [Google Scholar] [CrossRef] [PubMed]
- Erro, R.; Picillo, M.; Vitale, C.; Palladino, R.; Amboni, M.; Moccia, M.; Pellecchia, M.T.; Barone, P. Clinical clusters and dopaminergic dysfunction in de-novo Parkinson disease. Parkinsonism Relat. Disord. 2016, 28, 137–140. [Google Scholar] [CrossRef] [PubMed]
Mean ± SD | |
---|---|
Gender (female/male) | 9/17 |
Age in years | 71.19 ± 9.70 |
Age at diagnosis (years) | 65.77 ± 10.37 |
Visit 1 MDS-UPDRS-III | 28.89 ± 14.07 |
Visit 2 MDS-UPDRS-III | 24.12 ± 13.50 |
Visit 1 H&Y stage | 1.77 ± 0.71 |
Visit 2 H&Y stage | 1.85 ± 0.78 |
Visit 1 levodopa equivalent dose (mg) | 498.94 ± 309.88 |
Visit 2 levodopa equivalent dose (mg) | 637.40 ± 322.37 |
Time between clinical visits (days) | 65.62 ± 26.46 |
Cluster A | Cluster B | Cluster C | Cluster D | |
---|---|---|---|---|
Visit 2 MDS-UPDRS-III & PKG Scores | ||||
Study age (years) | 74.31 ± 9.99 | 67.96 ± 9.70 | 68.15 ± 3.96 | 68.29 ± 5.68 |
Age at diagnosis (years) | 69.69 ± 9.22 | 65.41 ± 9.08 | 60.48 ± 10.32 | 54.95 ± 4.15 † |
Years of PD | 4.62 ± 3.54 | 2.55 ± 1.78 | 7.67 ± 6.65 | 13.33 ± 2.05 † |
Gender (female/male)) | 5/8 | 2/5 | 1/2 | 1/2 |
Number of participants | 13 | 7 | 3 | 3 |
Best MDS-UPDRS-III | ||||
Study age (years) | 75.62 ± 7.84 | 65.59 ± 10.86 | 63.54 ± 0.00 | 67.40 ± 3.74 †† |
Age at diagnosis (years) | 70.26 ± 7.67 | 62.90 ± 10.35 | 46.54 ± 0.00 | 59.90 ± 8.17 |
Years of PD | 5.36 ± 4.55 | 2.69 ± 1.77 | 17.00 ± 0.00 | 7.50 ± 4.61 |
Gender (female/male) | 5/9 | 2/5 | 0/1 | 2/4 |
Number of participants | 14 | 7 | 1 | 4 |
Best PKG Scores | ||||
Study age (years) | 72.92 ± 10.21 | 67.66 ± 7.54 | 63.54 ± 0.00 | 70.93 ± 5.24 |
Age at diagnosis (years) | 68.69 ± 9.47 | 63.49 ± 8.78 | 46.54 ± 0.00 | 57.43 ± 2.74 |
Years of PD | 4.23 ± 3.31 | 4.17 ± 4.14 | 17.00 ± 0.00 | 13.50 ± 2.50 † |
Gender (female/male) | 6/11 | 3/3 | 0/1 | 0/2 |
Number of participants | 17 | 6 | 1 | 2 |
Cluster A | Cluster B | Cluster C | Cluster D | |
---|---|---|---|---|
Best MDS-UPDRS-III | ||||
LEDD | 387 ± 151 | 643 ± 127 | 1380 ± 0 | 1157 ± 183 |
Carbidopa/levodopa IR | 279 ± 159 | 629 ± 150 | 1050 ± 0 | 925 ± 299 |
Carbidopa/levodopa CR | 21 ± 80 | -- | 200 ± 0 | 300 ± 476 |
Ropinirole | -- | 1 ± 4 | 4 ± 0 | -- |
Selegiline | -- | -- | 10 ± 0 | 1 ± 1 |
Rasagiline | 0.2 ± 0.4 | -- | -- | -- |
Rytary | 194 ± 547 | -- | -- | -- |
Best PKG Scores | ||||
LEDD | 381 ± 105 | 942 ± 233 | 1380 ± 0 | 1131 ± 206 |
Carbidopa/levodopa IR | 335 ± 147 | 917 ± 183 | 1050 ± 0 | 250 ± 354 |
Carbidopa/levodopa CR | 18 ± 73 | 33 ± 82 | 200 ± 0 | 500 ± 707 |
Ropinirole | -- | -- | 4 ± 0 | -- |
Selegiline | 0.6 ± 2.4 | -- | 10 ± 0 | 1 ± 2 |
Rasagiline | 0.1 ± 0.3 | -- | -- | 0.3 ± 0.4 |
Rytary | 45 ± 184 | -- | -- | 1170 ± 1655 |
Demographics Alone | Demographics and MDS-UPDRS-III | Demographics and PKG | Demographics, MDS-UPDRS-III and PKG | |
---|---|---|---|---|
Sensitivity | 61.3 ± 1.0% | 65.2 ± 0.8% | 84.5 ± 0.7% | 86.5 ± 0.5% |
Specificity | 62.3 ± 1.5% | 66.0 ± 0.7% | 81.7 ± 2.2% | 87.7 ± 1.6% |
Accuracy | 61.6 ± 0.8% | 65.4 ± 0.6% | 83.8 ± 0.7% | 86.9 ± 0.6% |
PPV | 82.2 ± 0.6% | 84.4 ± 0.3% | 93.1 ± 0.8% | 95.3 ± 0.6% |
F1 Score | 70.1 ± 0.8% | 73.5 ± 0.6% | 88.5 ± 0.5% | 90.7 ± 0.4% |
AUC | 0.618 ± 0.008 | 0.656 ± 0.005 | 0.831 ± 0.011 | 0.871 ± 0.008 |
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |
© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Watts, J.; Khojandi, A.; Vasudevan, R.; Nahab, F.B.; Ramdhani, R.A. Improving Medication Regimen Recommendation for Parkinson’s Disease Using Sensor Technology. Sensors 2021, 21, 3553. https://doi.org/10.3390/s21103553
Watts J, Khojandi A, Vasudevan R, Nahab FB, Ramdhani RA. Improving Medication Regimen Recommendation for Parkinson’s Disease Using Sensor Technology. Sensors. 2021; 21(10):3553. https://doi.org/10.3390/s21103553
Chicago/Turabian StyleWatts, Jeremy, Anahita Khojandi, Rama Vasudevan, Fatta B. Nahab, and Ritesh A. Ramdhani. 2021. "Improving Medication Regimen Recommendation for Parkinson’s Disease Using Sensor Technology" Sensors 21, no. 10: 3553. https://doi.org/10.3390/s21103553
APA StyleWatts, J., Khojandi, A., Vasudevan, R., Nahab, F. B., & Ramdhani, R. A. (2021). Improving Medication Regimen Recommendation for Parkinson’s Disease Using Sensor Technology. Sensors, 21(10), 3553. https://doi.org/10.3390/s21103553