Data and Patients ‣ There are 44 items in parts I - III of UPDRS scale, each of which is rated between 0 (normal) and 4 (severely affected). ‣ Either (a) “total” score calculated as sum of parts I - III or (b) just one individual part is commonly used clinical endpoint in PD trial analyses. ‣ This work was performed using the UPDRS data from two clinical trials: • 12-week crossover Phase III clinical study in 161 early PD patients treated with at least one dose of controlled- and/or extended-release formulations of ropinirole [4]. • 24-week parallel-group Phase III clinical study in 393 advanced PD patients that were inadequately controlled by L-Dopa and treated with at least one dose of placebo or ropinirole [5] ‣ The same structural model (excluding the longitudinal aspects) developed previously was assumed and all the item specific parameters were fixed to their final estimates ‣ In the current work, baseline IRM was developed using UPDRS assessments before the start of the treatment (baseline) from these studies by adapting the previously developed model (based on MDS-UPDRS data) Item Response Theory modelling to leverage data from historical Parkinson’s Disease trials while integrating data from a newer version of the clinical endpoint Gopichand Gottipati 1 , Alienor C Berges 2 , Shuying Yang 2 , Chao Chen 2 , Mats O Karlsson 1 , Elodie L Plan 1 To develop a framework that can leverage historical Unified Parkinson’s Disease (PD) Rating Scale (UPDRS) clinical data, while integrating with a newer version - Movement Disorder Society sponsored revision (MDS-UPDRS) through the application of Item Response Theory (IRT) methodology Objectives Methods Results Background Conclusions 1 Dept of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden 2 GlaxoSmithKline, London, UK References [1] Baker FB, 2001 [2] PPMI database; www.ppmi-info.org [3] Gottipati G et al., PAGE 24 (2015) Abstr 3596 [4] Stocchi F et al., Curr Med Res Opin. 24:2883-95 2008 [5] Pahwa R et al al., Neurology. 68:1108-15, 2007 [6] Goetz C et al., Mov Disord 23: 2129-70, 2008 Disclosures: Uppsala University, the institution of G. Gottipati, M.O. Karlsson, and E. L. Plan, has received contract analysis fee from GlaxoSmithKline, London, UK A. C. Berges, S. Yang, C. Chen are employees of GlaxoSmithKline, London, UK. • IRT methodology allowed for integration of two versions of main clinical endpoints used in PD trial analyses into one unique framework, provided they are mapped on to the same underlying latent variable(s) continuum. • This can facilitate improved utilization of data from diverse sources and also different disease populations (e.g., early vs. advanced PD), potentially leading to better characterization of disease progression and drug effects in PD. ‣ Parameters characterising the re-assignment of the probabilities for the indirectly mapped items and the parameters for items unique to UPDRS were estimated successfully. ‣ The estimates for mean and variance for each of the latent variable distribution is shown in Table - 1 and plotted in Figure 3 (each panel representing a latent variable). ‣ The main assumption in IRT is that the probability of response (0, 1, …k, .. K) to each item (j) depends on an unobserved variable ‘disability’ of each subject (i) [1]. A two parameter logistic model was used to characterise this relationship (Eq. 1 & 2). ‣ Item Response Model (IRM) parameters are classified into: • Item specific parameters: discrimination/slope (a j ) and difficulty/location (b j ), are modelled as fixed effects • Subject specific parameter: disability (D i ), is modelled as random effects ‣ A longitudinal IRM using MDS-UPDRS data of De Novo cohort from the Parkinson’s Progression Markers Initiative database [2] was developed previously [3]: • It included three latent variables, whose assignment and choice was in good agreement with how the questionnaire was set up and also based on model-based diagnostics (Eq.1) (Eq.2) …. (Eq.1) …. (Eq.2) Step 1 Mapping - strategy outlined by Goetz et al [6] (Fig. 1 & 2) 1, 2, 6 - 8, 12, 15, 18, 22 (5)*, 23 (2)*, 24 (2)* 25 (2)* , 26 (2)* 28-30 3 - 5, 9 - 11, 16, 17, 19, 27, 31 13, 14, 20 (5)*, 21 (2)* Figure 1. Mapping mechanisms. Listed numbers are the items in UPDRS scale; highlighted in bold are items which were evaluated for different parts and/or sides (e.g., left, right) of the body and number of these evaluations are shown in parentheses with * Step 2 Estimation of the mean & variance of latent variable distributions with shifts Step 3 Model evaluation: simulation based diagnostics • The mean and variance of the latent variable distributions for the early PD subjects (used as the reference population), and the variances, shifts in the means for the study with advanced PD subjects were estimated Figure 2. Mapping strategy for item 29. Suggested scheme mapped from UPDRS to MDS - UPDRS: 0 - 0; 1 - 1, 2 - 2, 3 - 2, 4 - 3 or 4; indicated in bold are two additional parameters (FR2, FR4) that are estimated based on the adapted model UPDRS Data DV = 0 (U) P0 DV = 1 (U) P1 DV = 2 (U) P2 DV = 3 (U) P3 DV = 4 (U) P4 MDS - UPDRS Model DV = 0 (MD) P0 DV = 1 (MD) P1 DV = 2 (MD) P2 DV = 3 (MD) P3 DV = 4 (MD) P4 Adapted Final Model DV = 0 P0 DV = 1 P1 DV = 2 P2*FR2 DV = 3 P2*(1-FR2) + P4*FR4 DV = 4 P3 + P4*(1-FR4) Parameter Values for Study with Early PD Subjects Shift value for Study with Late PD Subjects Mean (Variance) Latent variable 1: Patient reported items (1 - 17) 0.535 (0.774) + 0.697 = 1.23 (1.88) Latent variable 2: Non-dexterous items (18, 19, 20*, 22*, 27 - 31) 0.219 (1.23) + 1.36 = 1.58 (2.63) Latent variable 3: Dexterous items (20*,21, 22*,23 - 26) 0.353 (0.821) + 0.32 = 0.734 (1.50) Table 1. Mean & variances of latent variable distributions. Number following ‘+’ is the shift; Items 20* and 22* were evaluated for tremor at rest of face, lips, chin and rigidity of neck respectively, among the right and left sides of the body and therefore they appear in both the non-dexterous as well as dexterous latent variable Figure 4. Simulation based diagnostics. The red line is the calculated mean of all the observed scores for each item; black histogram represent the distribution of the mean of simulated scores (based on 100 simulations of the final baseline model) and purple shaded area shows 5th and 97.5th percentiles for each item based on the the simulations; U/LE - Upper/Lower Extremity. Panels show the stratification based on the mapping mechanisms shown in Figure 1. ‣ The simulation based diagnostics (Figure. 4) suggest: • very good agreement with observations for items unique to UPDRS • varying degree of agreement for direct/indirectly-mapped items ‣ Although not all the IRM parameters were re-estimated, which allowed to use prior information and gain run-times, adequacy between the data and simulations was satisfactory. Figure 3. Distribution of the latent variables with shifts. Patient reported latent variable Non-dexterous latent variable Dexterous latent variable Model Adaptation Workflow • Direct-mapping: no additional parameter estimation was necessary • Indirect-mapping: for the items which do not map one-to-one, additional parameters reflecting the re-assignment of the probabilities were estimated (E.g., Item 29 in Fig. 2) • Only in UPDRS: item-specific parameters were estimated based on UPDRS data only as these items showed little to no-parallelism with MDS - UPDRS scale 1.Mentation 2.Thought Disorder 6.Salivation 7.Swallowing 8.Hand−Writing 12.Turning In Bed 15.Walking 18.Speech 22.Rigidity Neck 22.Rigidity UE (Left) 22.Rigidity UE (Right) 22.Rigidity LE (Left) 22.Rigidity LE (Right) 23.Finger Taps (Left) 23.Finger Taps (Right) 24.Hand movement (Left) 24.Hand movement (Right) 25.Hand Pronation Supination (Left) 25.Hand Pronation Supination (Right) 26.Leg Agility (Left) 26.Leg Agility (Right) 28.Posture 29.Gait 30.Postural Stability 0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60 1 2 1 2 1 2 1 2 Mean score count 3.Depression 4.Motivation/ Initiative 5.Speech 9.Cutting Food 10.Dressing 11.Hygeine 16.Tremor 17.Sensory Symptoms 19.Facial Expression 27.Arising From Chair 31.Bradykinesia Hypokinesia 0 20 40 60 0 20 40 60 0 20 40 60 0 20 40 60 1 2 1 2 Mean score count 13.Falling 14.Freezing 20.Tremor at Rest Face,Lips,Chin 20.Tremor at Rest Left Hand 20.Tremor at Rest Right Hand 20.Tremor at Rest Left Foot 20.Tremor at Rest Right Foot 21.Action/Postural Tremor Left Hand 21.Action/Postural Tremor Right Hand 0 20 40 60 0 20 40 60 0 20 40 60 1 1 1 Mean score count
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Data and Patients ‣ There are 44 items in parts I - III of UPDRS scale, each of which is rated between
0 (normal) and 4 (severely affected). ‣ Either (a) “total” score calculated as sum of parts I - III or (b) just one individual part is
commonly used clinical endpoint in PD trial analyses. ‣ This work was performed using the UPDRS data from two clinical trials:
• 12-week crossover Phase III clinical study in 161 early PD patients treated with at least one dose of controlled- and/or extended-release formulations of ropinirole [4].
• 24-week parallel-group Phase III clinical study in 393 advanced PD patients that were inadequately controlled by L-Dopa and treated with at least one dose of placebo or ropinirole [5]
‣ The same structural model (excluding the longitudinal aspects) developed previously was assumed and all the item specific parameters were fixed to their final estimates
‣ In the current work, baseline IRM was developed using UPDRS assessments before the start of the treatment (baseline) from these studies by adapting the previously developed model (based on MDS-UPDRS data)
Item Response Theory modelling to leverage data from historical Parkinson’s Disease trials while integrating data from
a newer version of the clinical endpoint Gopichand Gottipati1, Alienor C Berges2, Shuying Yang2, Chao Chen2, Mats O Karlsson1, Elodie L Plan1
To develop a framework that can leverage historical Unified Parkinson’s Disease (PD) Rating Scale (UPDRS) clinical data, while integrating with a newer version - Movement Disorder Society sponsored revision (MDS-UPDRS) through the application of Item Response Theory (IRT) methodology
Objectives
Methods
Results
Background
Conclusions
1 Dept of Pharmaceutical Biosciences, Uppsala University, Uppsala, Sweden 2 GlaxoSmithKline, London, UK
References[1] Baker FB, 2001
[2] PPMI database; www.ppmi-info.org
[3] Gottipati G et al., PAGE 24 (2015) Abstr 3596
[4] Stocchi F et al., Curr Med Res Opin. 24:2883-95 2008
[5] Pahwa R et al al., Neurology. 68:1108-15, 2007
[6] Goetz C et al., Mov Disord 23: 2129-70, 2008
Disclosures: Uppsala University, the institution of G. Gottipati, M.O. Karlsson, and E. L. Plan, has received contract analysis fee from GlaxoSmithKline, London, UK A. C. Berges, S. Yang, C. Chen are employees of GlaxoSmithKline, London, UK.
• IRT methodology allowed for integration of two versions of main clinical endpoints used in PD trial analyses into one unique framework, provided they are mapped on to the same underlying latent variable(s) continuum.
• This can facilitate improved utilization of data from diverse sources and also different disease populations (e.g., early vs. advanced PD), potentially leading to better characterization of disease progression and drug effects in PD.
‣ Parameters characterising the re-assignment of the probabilities for the indirectly mapped items and the parameters for items unique to UPDRS were estimated successfully.
‣ The estimates for mean and variance for each of the latent variable distribution is shown in Table - 1 and plotted in Figure 3 (each panel representing a latent variable).
!!!!!!!!!
!!!!!!
!!!!!!
‣ The main assumption in IRT is that the probability of response (0, 1, …k, .. K) to each item (j) depends on an unobserved variable ‘disability’ of each subject (i) [1]. A two parameter logistic model was used to characterise this relationship (Eq. 1 & 2).
‣ Item Response Model (IRM) parameters are classified into: • Item specific parameters: discrimination/slope (aj) and difficulty/location (bj), are modelled
as fixed effects • Subject specific parameter: disability (Di), is modelled as random effects !!!!
‣ A longitudinal IRM using MDS-UPDRS data of De Novo cohort from the Parkinson’s Progression Markers Initiative database [2] was developed previously [3]: • It included three latent variables, whose assignment and choice was in good agreement
with how the questionnaire was set up and also based on model-based diagnostics
(Eq.1)
(Eq.2)
…. (Eq.1)
…. (Eq.2)
Step 1 Mapping - strategy outlined by Goetz et al [6] (Fig. 1 & 2)
1, 2, 6 - 8, 12, 15, 18, 22 (5)*, 23 (2)*, 24 (2)*
25 (2)* , 26 (2)* 28-30
3 - 5, 9 - 11, 16, 17, 19, 27, 31
13, 14, 20 (5)*, 21 (2)*
Figure 1. Mapping mechanisms. Listed numbers are the items in UPDRS scale; highlighted in bold are items which were evaluated for different parts and/or sides (e.g., left, right) of the body and number of these evaluations are shown in parentheses with *
Step 2 Estimation of the mean & variance of latent variable distributions with shifts
Step 3 Model evaluation: simulation based diagnostics
• The mean and variance of the latent variable distributions for the early PD subjects (used as the reference population), and the variances, shifts in the means for the study with advanced PD subjects were estimated
Figure 2. Mapping strategy for item 29. Suggested scheme mapped from UPDRS to MDS - UPDRS: 0 - 0; 1 - 1, 2 - 2, 3 - 2, 4 - 3 or 4; indicated in bold are two additional parameters (FR2, FR4) that are estimated based on the adapted model
UPDRS Data DV = 0 (U) P0
DV = 1 (U) P1
DV = 2 (U) P2
DV = 3 (U) P3
DV = 4 (U) P4
MDS - UPDRS Model DV = 0 (MD) P0
DV = 1 (MD) P1
DV = 2 (MD) P2
DV = 3 (MD) P3
DV = 4 (MD) P4
Adapted Final Model DV = 0 P0
DV = 1 P1
DV = 2 P2*FR2
DV = 3 P2*(1-FR2) + P4*FR4
DV = 4 P3 +
P4*(1-FR4)
Parameter
Values for Study with Early PD Subjects
Shift value for Study with Late PD Subjects
Mean (Variance)
Latent variable 1: Patient reported items (1 - 17)
0.535 (0.774)
+ 0.697 = 1.23 (1.88)
Latent variable 2: Non-dexterous items (18, 19, 20*, 22*, 27 - 31)
0.219 (1.23)
+ 1.36 = 1.58 (2.63)
Latent variable 3: Dexterous items (20*,21, 22*,23 - 26)
0.353 (0.821)
+ 0.32 = 0.734 (1.50)
Table 1. Mean & variances of latent variable distributions. Number following ‘+’ is the shift; Items 20* and 22* were evaluated for tremor at rest of face, lips, chin and rigidity of neck respectively, among the right and left sides of the body and therefore they appear in both the non-dexterous as well as dexterous latent variable
Figure 4. Simulation based diagnostics. The red line is the calculated mean of all the observed scores for each item; black histogram represent the distribution of the mean of simulated scores (based on 100 simulations of the final baseline model) and purple shaded area shows 5th and 97.5th percentiles for each item based on the the simulations; U/LE - Upper/Lower Extremity. Panels show the stratification based on the mapping mechanisms shown in Figure 1.
‣ The simulation based diagnostics (Figure. 4) suggest: • very good agreement with observations for items unique to UPDRS • varying degree of agreement for direct/indirectly-mapped items
‣ Although not all the IRM parameters were re-estimated, which allowed to use prior information and gain run-times, adequacy between the data and simulations was satisfactory.
Figure 3. Distribution of the latent variables with shifts. Patient reported latent variable Non-dexterous latent variable Dexterous latent variable
Model Adaptation Workflow !!
• Direct-mapping: no additional parameter estimation was necessary • Indirect-mapping: for the items which do not map one-to-one, additional parameters
reflecting the re-assignment of the probabilities were estimated (E.g., Item 29 in Fig. 2) • Only in UPDRS: item-specific parameters were estimated based on UPDRS data only as
these items showed little to no-parallelism with MDS - UPDRS scale 1.Mentation 2.ThoughtDisorder 6.Salivation 7.Swallowing 8.Hand−Writing
12.TurningIn Bed 15.Walking 18.Speech 22.Rigidity
Neck22.RigidityUE (Left)
22.RigidityUE (Right)
22.RigidityLE (Left)
22.RigidityLE (Right)
23.Finger Taps(Left)
23.Finger Taps(Right)
24.Hand movement(Left)
24.Hand movement(Right)
25.Hand PronationSupination (Left)
25.Hand PronationSupination (Right)
26.Leg Agility(Left)
26.Leg Agility(Right) 28.Posture 29.Gait 30.Postural Stability
0204060
0204060
0204060
0204060
0204060
1 2 1 2 1 2 1 2Mean score
coun
t
3.Depression 4.Motivation/Initiative 5.Speech
9.Cutting Food 10.Dressing 11.Hygeine
16.Tremor 17.SensorySymptoms
19.FacialExpression
27.ArisingFrom Chair
31.BradykinesiaHypokinesia
0204060
0204060
0204060
0204060
1 2 1 2Mean score
coun
t
13.Falling 14.Freezing 20.Tremor at RestFace,Lips,Chin
20.Tremor at RestLeft Hand
20.Tremor at RestRight Hand
20.Tremor at RestLeft Foot
20.Tremor at RestRight Foot
21.Action/PosturalTremor Left Hand
21.Action/PosturalTremor Right Hand
0
20
40
60
0
20
40
60
0
20
40
60
1 1 1Mean score
coun
t
Related Documents
