Table 2 Statistical significance of variables used in disease state definition (i.e., clinical assessments) and action (medications).

From: Computational medication regimen for Parkinson’s disease using reinforcement learning

Time

Variables

Coef

p-value

\({t}_{1}\)

(Variables used in disease state definition)

  

Hoehn and Yahr

0.8967

0

Age

0.0864

0

Total MoCA

− 0.0534

0.01

Total UPDRS III scores

0.7816

0

% change of total UPDRS scores

− 0.0246

0

Subtype AR (vs. mixed type)

1.5576

0

Subtype TD (vs. mixed type)

0.6056

0.213

Levodopa

5.5486

0

Dopamine agonist

4.5439

0

Other medicine

2.2017

0

Levodopa + Other medicine

5.942

0

Levodopa + Dopamine agonist

5.5351

0

Dopamine agonist + Other medicine

3.8986

0

Levodopa + Dopamine agonist + Other medicine

6.1054

0

\({t}_{2}\)

(Variables used in actions)

  

Levodopa

− 6.9801

0

Dopamine agonist

− 5.0293

0

Other medicine

− 0.825

0.094

Levodopa + other medicine

− 6.9232

0

Levodopa + Dopamine agonist

− 7.2501

0

Dopamine agonist + other medicine

− 5.1182

0

Levodopa + Dopamine agonist + other medicine

− 8.5083

0

(Outcome) total UPDRS III scores

  
  1. We fitted prediction model as total UPDRS scores at \({t}_{2}\) (penalty) ~ variables at \({t}_{1}\) (used in disease state) + variables at \({t}_{2}\) (actions) using multivariate regression (i.e., generalized least square) and selected variables that are statistically significant on total UPDRS scores on the next timestep (\({t}_{2}\)). Note that the % change of total UPDRS scores are 100% \(\times\) (Total UPDRS scores at \({t}_{1}\) − Total UPDRS scores at \({t}_{0}\))/Total UPDRS scores at \({t}_{0}\). AR = Akinetic-rigid, TD = tremor-dominant.