Deep Brain Stimulation for Parkinson’s Disease with Early Motor Complications: A UK Cost-Effectiveness Analysis

Background Parkinson’s disease (PD) is a debilitating illness associated with considerable impairment of quality of life and substantial costs to health care systems. Deep brain stimulation (DBS) is an established surgical treatment option for some patients with advanced PD. The EARLYSTIM trial has recently demonstrated its clinical benefit also in patients with early motor complications. We sought to evaluate the cost-effectiveness of DBS, compared to best medical therapy (BMT), among PD patients with early onset of motor complications, from a United Kingdom (UK) payer perspective. Methods We developed a Markov model to represent the progression of PD as rated using the Unified Parkinson's Disease Rating Scale (UPDRS) over time in patients with early PD. Evidence sources were a systematic review of clinical evidence; data from the EARLYSTIM study; and a UK Clinical Practice Research Datalink (CPRD) dataset including DBS patients. A mapping algorithm was developed to generate utility values based on UPDRS data for each intervention. The cost-effectiveness was expressed as the incremental cost per quality-adjusted life-year (QALY). One-way and probabilistic sensitivity analyses were undertaken to explore the effect of parameter uncertainty. Results Over a 15-year time horizon, DBS was predicted to lead to additional mean cost per patient of £26,799 compared with BMT (£73,077/patient versus £46,278/patient) and an additional mean 1.35 QALYs (6.69 QALYs versus 5.35 QALYs), resulting in an incremental cost-effectiveness ratio of £19,887 per QALY gained with a 99% probability of DBS being cost-effective at a threshold of £30,000/QALY. One-way sensitivity analyses suggested that the results were not significantly impacted by plausible changes in the input parameter values. Conclusion These results indicate that DBS is a cost-effective intervention in PD patients with early motor complications when compared with existing interventions, offering additional health benefits at acceptable incremental cost. This supports the extended use of DBS among patients with early onset of motor complications.


APPENDIX A. SYSTEMATIC REVIEWS
The following sections present detailed information on the systematic reviews of evidence that informed the economic analysis.
At all stages of each review the study selection was performed independently by two analysts.
Verification at the level of abstracts and titles was carried out in such a way that all reports deemed useful by at least one of the analysts were enrolled to the next stage. In case of disagreements during the verification based on the full-text publications, the final decision on inclusion or exclusion was agreed by discussion and consensus.
Extraction of data from the studies included in the analysis was carried out by one of the analysts.
Then the extraction was verified by a second analyst.

Search strategy
In order to identify data on effectiveness and safety of CSAI and CDLCI a systematic search was performed. Due to the fact that search was performed at the initial phase of the project, when model structure and data included in the analysis were unknown, broad inclusion criteria were defined (details in Table 4). Once the model structure was decided, selection criteria were narrowed to include only studies reporting CSAI/CDLCI effectiveness measured in terms of UPDRS scores.
Additionally, studies reporting data on safety, withdrawal and CSAI dose were selected from the studies initially included in full text analysis.
The following databases were searched for papers published from the inception of each database to 2015: • Medline (via PubMed), • Embase, • The Cochrane Library. The search was performed on 11 th February 2015. Search strategies are outlined in Table 1, Table 2  and Table 3.

Studies selection
All studies identified within search process were initially assessed based on title and abstracts. Papers not meeting predefined inclusion criteria as indicated by title and abstracts were excluded.
In Table 4 below selection criteria defined according to PICO scheme are shown. • Cross-over studies, if no results were presented before patients cross, • Follow-up < 1 week, • Case series or studies with < 10 patients.

Other
• Primary studies, • Publications available in full-text, • Publications in English.
• Reviews, systematic reviews, meta-analysis, • Comments, editorials, letters, • Conference abstracts (unless concern results of primary study independently identified in search).
Publication date • Not relevant (up to date of the search -11 th February 2015) -Forty one studies (described in 49 publications) were included after full text analysis based on selection criteria. Additionally for 3 of finally included studies datasets on clinicaltrials.gov website for 3 studies were found and analysed. [1][2][3][4][5][6].
For the purposes of the economic analysis, additional selection was done and studies were excluded from final calculations of effectiveness due to one of the following reasons: • no data on UPDRS or UPDRS not measured in ON-meds condition, • insufficient data on UPDRS (only at baseline or no baseline values), • results presented as medians, • results presented at non-specified time points, e.g. after mean or median follow-up.
Studies were also included in final analysis if any of the following data was reported: CSAI mean daily dose, frequency of adverse events (specifically skin nodules / skin reaction for CSAI and peritonitis for CDLCI) and withdrawal rates.
A brief summary of all 3 searches and results of studies selection is presented in the table below. The sequential phases of the review process are also illustrated in the PRISMA flow diagram (Figure 1).   Table 6 and Table 7

CSAI -effectiveness (UPDRS)
Studies included in the analysis for particular UPDRS scores are listed in Table 8. A brief description of the studies is provided in Table 9.  calculated for the purposes of the economic analysis are presented in Table 10.

A.1.3.2 CSAI -other parameters
Studies included in the analysis for particular data categories are listed in Table 11.

Withdrawal data
The data extracted from the studies included in the analysis is presented in Table 12. Detailed information on final model assumptions is presented in section B.4.1.

Adverse events data
Data on skin reaction / skin nodules frequency was retrieved from the studies identified in the systematic review, as these were the most common adverse event for CSAI treatment reported in the literature. Other adverse events reported in the literature were not included in the model, as they were considered too rare (occurring in less than 5% of patients) or not influencing patient's quality of life or cost-generating.
The data extracted from the studies included in the analysis (only studies reporting frequency of skin nodules / skin reaction) is presented in Table 13. Detailed information on final model assumptions is presented in section B.3.1.

Dosage data
The data extracted from the studies included in the analysis is presented in Table 14. Detailed information on final model assumptions is presented in section B.5.

A.1.3.3 CDLCI -effectiveness
Studies included in the analysis for particular UPDRS scores are listed in Table 15. Brief description of the studies is provided in Table 16. for the purposes of the economic analysis are presented in Table 17. a) number of patients not reported, assumed the same as number of patients that completed 1-year follow-up; b) patients randomized to CDLCI at baseline; c) patients randomized to BMT at baseline who started CDLCI after 12 weeks Detailed information on further data analysis and final model assumptions are presented in section B.1.2.

A.1.3.4 CDLCI -other parameters
Studies included in the analysis for particular data categories are listed in Table 18.

Withdrawal data
The data extracted from the studies included in the analysis is presented in Table 19. Detailed information on final model assumptions is presented in section B.4.2.

Adverse events data
Data on frequency of peritonitis was retrieved from the systematic review, presented in Table 20.
Detailed information on final model assumptions is presented in section B.3.2.

Search strategy
In order to identify data on long-term effectiveness of DBS a systematic literature search was performed. Due to the fact that search was performed at the initial phase of the project, when model structure and data included in the analysis were unknown, broad inclusion criteria were defined (details in Table 24). Once the model structure was decided, selection criteria were narrowed to include only studies reporting long-term DBS effectiveness measured in terms of UPDRS scores.
The following databases were searched for papers published from the inception of each database to 2015: • Medline (via PubMed), • Embase, • The Cochrane Library.
The search was performed on 10 th March 2015. Search strategies are outlined in Table 21, Table 22 and Table 23. ((((Control or prospective or comparative* OR cohort* OR "cohort studies" OR (cohort and (study OR studies)) OR (cohort analy*) OR retrospectiv* OR observational OR (observational AND (study OR studies)) OR "cross-sectional" OR "cross sectional" OR "cross sectional" OR "cross-sectional studies" OR "case control studies" OR "Case control" OR ("follow up" and (study OR studies)) OR longitudinal Or retrospective OR (clinical AND (study OR trial)) OR crossover OR "cross-over" OR population-based OR survey OR register OR registry OR "case series")))) 7601004

#12 AND #15 2190
Search strategy executed on 10 th March 2015 control OR prospective OR comparative* OR cohort* OR 'cohort studies' OR (cohort AND (study OR studies)) OR (cohort AND analy*) OR retrospectiv* OR observational OR (observational AND (study OR studies)) OR 'cross-sectional' OR 'cross sectional' OR 'cross-sectional studies' OR 'case control studies' OR 'case control' OR ('follow up' AND (study OR studies)) OR longitudinal OR retrospective OR (clinical AND (study OR trial)) OR crossover OR 'cross-over' OR 'population based' OR survey OR register OR registry OR 'case series' 9,546,035

14
'randomized controlled trial' OR random* OR rct OR (singl* OR doubl* OR trebl* OR tripl* AND (blind* OR mask*)) OR (single AND blind) OR (double AND blind) OR (triple AND blind) OR placebo OR 'placebo controlled' OR blinding OR (controlled AND clinical AND trial) OR (random* AND controlled AND study*) OR (random* AND controlled AND trial*) OR (random OR randomly AND (allocation OR allocate*))

A.2.2. Study selection
All studies identified in the search were initially assessed based on title and abstracts applying predefined selection criteria listed below (Table 24). In a second step, publications initially included based on title and abstract screening, were assessed based on the full article. Papers not meeting the inclusion criteria were excluded from further evaluation.
In the table below selection criteria defined according to PICO scheme are shown. • Cross-over studies, if no results were presented before patients cross, • Case series or studies with < 20 patients.

Other
• Primary studies, • Publications available in full-text, • Publications in English.
• Reviews, systematic reviews, meta-analysis, • Comments, editorials, letters, • Conference abstracts (unless concern results of primary study independently identified in search), • Data not allowing to calculate change per given time period.
Publication date • Not relevant (up to date of the search -10 th March

2015) -
For the purposes of economic analysis, additional selection criteria were taken into account, and only studies meeting the following criteria were included: • Studies reporting UPDRS measured ON-meds, • Studies with results presented as mean (not median), • Studies reporting UPDRS scores at least at two follow-up time-points with first of them at least 2 year after treatment initiation.
A brief summary of all 3 searches and results of studies selection is presented below ( Table 25, Table   26). The sequential phases of the review process are also illustrated in the flow diagram ( Figure 2).
The list of studies included in final analysis together with data extracted from the studies are presented in section A.2.3.

Data extraction
Studies included in the analysis for particular UPDRS scores are listed in Table 27. Brief description of the studies is provided in Table 28.   The data extracted from the studies included in the analysis is presented below (Table 29, Table 30, Table 31, Table 32).  (52) 3.20 (52)

A.3.1. Search strategy
In order to identify data on long-term progression of UPDRS outcomes (for modelling of "natural disease progression") a systematic search of "BMT" studies was performed in Medline (via PubMed).
"BMT" was defined as any medical intervention (except DBS, CSAI, CDLCI), and intervention studies of PD medication, or population-based studies, were included. The search in Medline database was performed on 12 th February 2015. The search was performed in the initial phase of the project, when model structure and data included in the analysis were unknown. Therefore, broad inclusion criteria were defined (details in Table 34).
An additional non-systematic search via google and in ISPOR database [241] was done. As this was non-systematic no specific strategy was used.

A.3.2. Study selection
All studies identified within search process were initially assessed based on title and abstracts. Papers not meeting predefined inclusion criteria as indicated by title and abstracts were excluded.
A summary of primary selection criteria is presented in Table 34.   (Table 36). As a result, 30 additional papers were included.
For the purposes of economic analysis, additional selection was done and only studies reporting data on UPDRS progression were finally included. Additional selection criteria, in order to have data to derive UPDRS progression rates, were: • UPDRS scores measured ON-meds, • at least 2-year follow-up (excluding initial period of treatment -if specific intervention assessed, first 2 years of treatment excluded from the analysis -to eliminate short-term effectiveness), • studies performed in early / recently diagnosed PD patient were excluded (the exception is Reinosso 2014 study [242] -performed in recently diagnosed PD patients, but results reported separately for period 7-9 years from diagnosis).
A brief summary of the search and results of study selection is presented in Table 35 and Table 36.  Table 36 Summary of systematic search for PD progression in BMT treated patients -Additional search

Data source No. of considered studies (initial selection) Final no. of included studies
References of the retrieved publications 24 1 Other (google, ISPOR) 6 0

Figure 3. Overview of studies selection: PRISMA flowchart -long-term progression in BMT studies
No data for UPDRS IV progression were identified. The studies included in the final analysis for UPDRS I-III are listed in Table 37.

Lack of data specified in inclusion criteria (n = 49)
Studies included in the analysis -initial selection (n = 56) Studies included in the analysis -final selection (n = 6) The publications included after initial selection, but finally excluded from the analysis are listed in Table 38.

A.3.3. Data extraction
Brief description of the studies included in the analysis is provided in Table 39. The data extracted from the studies included in the analysis is presented in Table 40.

A.4.1. Search strategy
In order to identify data on mortality in PD patients Medline database (via PubMed) was systematically searched. Search was performed in two steps: 1. Firstly, other systematic reviews were identified; 2. Secondly, a systematic search was done in order to identify studies publicized after date of search in latest review identified in step 1.
Detailed search strategy for other systematic reviews is presented in Table 41. This search was done on 25 th March 2015. Out of 30 abstracts identified in the search, 2 publications (describing systematic reviews) were included in full text analysis [297,298]. The latest systematic review identified was performed by  Table   42. The date of this search was 23 th April 2015. Additionally, a non-systematic search via google was done.

A.4.2. Study selection
All studies identified within the Medline database search (studies published from 2012 onwards) were first analysed based on predefined selection criteria described in Table 43.  • Data allowing for determining relationship between mortality and progression assessed with one of following: H&Y, UPDRS.
• Data allowing to calculate mortality in general with PD.
• Lack of data specified in inclusion criteria.
• RCT studies, • Case series. • Publicized in English -A brief summary of additional search is presented in Table 44. Details on selection of studies included in MacLeod 2014 review are presented in Table 45. Additionally one study was identified in non-systematic search via google and this study was also included in the analysis [299].
The diagram below (Figure 4) provides information on selection process for: • 3 publications (out of 176) included in MacLeod 2014 review, that were selected for full text analysis based on information provided in MacLeod 2014 review, • 348 publications identified through additional search in Pubmed (via Medline), to identify studies published after MacLeod 2014., • 1 publication identified in non-systematic search via google. List of publications finally included in the analysis is presented in Table 46.

Total number of publications 3
List of publications included after initial selection, but finally excluded from the analysis is presented in Table 47. Table 47.

Studies on mortality not included in final calculation (included after initial selection)
Publication

Reason of exclusion
Jones 2012 [302] association between UPDRS and mortality not examined in the study

Hobson 2010 [309]
association between UPDRS and mortality not examined in the study

Hoyert 2012 [310]
association between UPDRS and mortality not examined in the study

Kaltenboeck 2012 [311]
association between UPDRS and mortality not examined in the study

Mackenbach 2014 [312]
association between UPDRS and mortality not examined in the study

Merola 2014 [149]
association between UPDRS and mortality not examined in the study

Morgan 2014 [313]
association between UPDRS and mortality not examined in the study

Oosterveld 2015 [314]
association between UPDRS and mortality examined, but reported results not useful for current analysis purposes

Peretz 2014 [315]
association between UPDRS and mortality not examined in the study

Vu 2012 [293]
association between UPDRS and mortality examined, but only total UPDRS or UPDRS II considered, whereas in the model only association between UPDRS III and mortality was taken into account (as most commonly reported)

A.4.3. Data extraction
Studies finally included in the analysis together with data extracted from the studies are presented in Table 48.

B.1. CSAI and CDLCI short-term effectiveness (first 2 years in the model)
The following sections present detailed information on the analysis of CSAI and CDLCI short-term effectiveness data (UPDRS) that was performed to derive values for short-term effectiveness parameters (first 2 years in the model) included in the CE model for these therapies. Data included in the analyses are from studies identified in the systematic review of both therapies (details in section A.1).

UPDRS I
The Drapier 2012 study indicates a 17% improvement with CSAI in part I of the UPDRS scale after 1 year of treatment vs baseline.  Table 50.   Table 52.

UPDRS III
CSAI effectiveness data on the UPDRS part III scale were reported in 3 studies -results after 1 year and 2 years of treatment are available Data on percentage changes in the UPDRS III vs baseline calculated from these studies are presented in Table 53.  [16], showing an 45% improvement in UPDRS motor score after 6 months of treatment up to 2 years in patients with average baseline UPDRS III score of 29.7, support the conclusion that higher effectiveness is reached in more advanced patients. As no information about UPDRS measurement condition (on or off meds) was provided in Kanovsky 2002, this study was excluded from the quantitative analysis.
Finally, in economic analysis only data from Todorowa 2013 were used. It was conservatively assumed that improvement in UPDRS III score is seen immediately after treatment initiation and is maintained for 2 years of treatment (year 2 data used to populate CE model). The final model assumptions are presented in Table 54.

UPDRS IV
CSAI effectiveness data on the UPDRS part IV scale were reported in 5 studies and data for 3 time points are available. Data on percentage changes in the UPDRS IV vs baseline calculated from these studies are presented in Table 55.  2. The Kanovsky 2002 study was the only study with more than one time point and so results of this study were used to calculate the % change at 2 years of treatment relative to year 1 -1.89% worsening after 2 years when compared to year 1 results (details in Table 56): 3. No data on initial effectiveness (immediately after treatment initiation) were identified. The results of 2 studies included in the analysis show that improvement in UPDRS IV is achieved as soon as after 6 months of treatment (see Table 55). It was conservatively assumed that initial effectiveness is the same as after first year of treatment.
The final model assumptions are presented in Table 57. year of treatment were available, for which percentage changes to baseline were calculated (Table   58).  In the economic analysis, the weighted average % change after first year of treatment vs baseline was calculated based on data from all studies reporting respective data (Table 58): 4.72%. It was conservatively assumed that UPDRS I score does not change immediately after treatment initiation.
Effectiveness after 2 years of treatment was assumed to be the same as after 1 year of treatment. The final model assumptions are presented in Table 59.

UPDRS II
CDLCI effectiveness data on the UPDRS part II scale were reported in 5 studies and data for 5 different time points are available. Data on percentage changes in the UPDRS II vs baseline calculated from these studies are presented in Table 60. 2. The Fernandez 2014 study results were used to calculate % change at 1 year of treatment relative to initial results -11.86% worsening after 1 year when compared to initial results (details in Table   61):  The final model assumptions are presented in Table 63.

UPDRS III
CDLCI effectiveness data on the UPDRS part III scale were reported in 10 studies (11 publications).
In 5 studies, CDLCI effectiveness was assessed at a single time point only. Data on percentage changes in the UPDRS III vs baseline calculated from all studies included in the analysis are presented in Table 64.  For the purposes of CE model, the effectiveness was estimated as follows: 1. The weighted average % change after 1 year of treatment vs baseline (the most commonly considered time point in studies included in the analysis) was calculated based on the results of 6 studies (7 groups of patients) reporting respective data --17.65%.
2. The Fernandez 2014 study results were used to calculate % change at 1 year of treatment relative to initial results (4 weeks results) -9.47% worsening after 1 year when compared to initial results (details in Table 65):   The final model assumptions are presented in Table 67.

UPDRS IV
CDLCI effectiveness data on the UPDRS part IV scale were reported in 6 studies and data for 3 time points are available. Data on percentage changes in the UPDRS IV vs baseline calculated from these studies are presented in Table 68. 2. The Antonini 2008 study results were used to calculate % change at 2 years of treatment relative to year 1 -3.12% worsening after 2 years when compared to year 1 results (details in Table 69): 3. It was conservatively assumed that initial % change vs baseline is the same as after first year of treatment (-44.61%).
The final model assumptions are presented in Table 70. used. However, all these data are not sufficient to estimate change in total part IV of UPDRS over time. Therefore, data from EARLYSTIM (BMT arm) were also considered for estimation of UPDRS IV progression in the model, and for the analysis to derive UPDRS I-III progression rates (see Table 71).

B.2.2. UPDRS long-term progression in DBS studies
Data on UPDRS long-term outcomes were also collected from DBS studies, in the systematic review described in section A.2. Only studies reporting UPDRS scores at least at two follow-up time-points with first of them at least 2 year after treatment initiation were included in the analysis (other selection criteria listed in section A.2.2). Data used in the analysis are summarised in Table 72, Table 73, Table   74, Table 75. In order to avoid using the same patient group data more than once, publications were collected into groups describing the same data and considered as one study.     (12) The data presented in the tables above were furtherly analyzed with use of linear regression (geeglm function from geepack package for R version 3.2.2) to derive long-term progression rates in patients treated with DBS (separately for each UPDRS score). Use of geeglm function allowed for estimation of single annual progression rate for each UPDRS score based on data from different groups of patients reported at multiple time points (not necessarily the same time points in all groups of patients), weighted with number of patients in each group that was changing over time (patients were lost to follow-up).  The reliability of the results presented above is limited.  Table 75) suggest that UPDRS IV score after 5 years of treatment is almost 2 points worse than after 2 years (5.10 at year 5 vs 3.20 at year 2). However, in this analysis, the observed worsening in UPDRS IV was not statistically significant and so the results do not allow a rejection of the hypothesis that UPDRS IV remains constant in the long term setting.

B.2.3. CE model assumptions
Results of the analysis of long-term UPDRS progression in DBS and BMT studies are summarized in  [128], one of the DBS studies included in the analysis, deemed to be representative for UPDRS III progression in population under consideration, based on clinical expert opinion. As mentioned before, long term evidence for CSAI and CDLCI was not identified.
Given the data availability, reliable assessment of possible changes in progression rates over time (depending on duration of disease, treatment, motor complication etc.) was not possible. Thus, long term UPDRS progression is expressed in terms of a constant annual progression rates.  Table 30 and Table 31) Based on clinical expert opinion, it was assumed that long term progression rates are the same, Due to high uncertainty in estimated progression rates alternative data sources were considered in the sensitivity analysis. Assumptions of base case scenario and sensitivity analysis scenarios are presented in Table 78. Additionally, scenarios assuming different response duration for DBS, CSAI and CDLCI were explored in the sensitivity analysis: • for UPDRS I-III: value at year 2 held constant (no progression) for a further 2 years, and the same progression rate as for BMT arm applied thereafter; • for UPDRS IV: the same progression rate as for BMT arm applied from year 3 onwards.

B.3. Adverse events
Data on adverse events associated with DBS, CSAI and CDLCI were identified in the systematic review (details in Appendix A).

B.3.1. CSAI -skin nodules frequency
Studies reporting data on skin nodules / skin reaction frequency during CSAI treatment were identified in systematic review (details in section A.1). The data from studies included in the analysis are presented in Table 79. Based on these data weighted mean frequency of skin nodules/ skin reaction occurrence was calculated (74.56%). Given the lack of long term data, this rate was applied only within the 1 st year of treatment. Data on peritonitis frequency during CDLCI treatment were retrieved from the studies identified in the systematic review, and the analysis of the data is presented in Table 80. Based on these data, a weighted mean frequency of peritonitis occurrence was calculated (3.23%). Given the lack of long term data, this rate was applied only within the 1 st year of treatment.

B.4. Withdrawal
Data on withdrawal rates for CSAI and CDLCI were sourced from studies identified in the systematic review of CSAI and CDLCI efficacy (details in section A.1).

B.4.1. CSAI
Detailed data on treatment discontinuation (including information on follow-up period) were available in 9 studies (12 publications). Data on withdrawal rates in the studies included in the analysis are presented in Table 81. Only withdrawals due to adverse events, lack of efficacy or due to own patient decision were included in the calculations. The probability of treatment discontinuation was calculated based on data from all studies listed above. In the first step, yearly probability of discontinuation was calculated for each study (probability from de Gaspari 2006 was calculated based on mean follow-up period, data on total follow-up period was not used), assuming constant yearly withdrawal. In the second step, the weighted mean withdrawal probability was estimated for all studies. Values of weights were assumed to be equal to the product of number of patients and follow-up period in each study. A summary of calculations is presented in Table 82. The weighted mean was 7.59% (SD 0.12%) and this value was used in the model.  Table 83. Discontinuation rates were estimated separately for initial phase / test phase of treatment, first year of treatment (initial phase excluded) and subsequent years of treatment, according to the information in the studies. Summary of discontinuation rates applied in the model is provided in Table 84. Mean daily dose of CSAI was calculated using data on mean dose after achieving optimal dosage or mean dose within study follow-up. In case data was available at several time points in the study, data from the last time point were used. This assumption has no effect on the results of the analysis as the differences in daily doses over time were not significant in such cases.
Detailed data from single studies used in calculation are presented below, together with mean daily dose weighted by number of patients.

B.6. Falls
No separate systematic review was performed to identify data on falls in PD patients. Relevant data were selected from studies identified in the search for progression data in BMT studies (details in section A.3). Initial selection was based on titles and abstracts and only studies specifically aimed at assessing falls in PD patients were included in full text analysis. Out of 15 publications included in full text analysis, 7 were finally included in the analysis (excluded studies did not provide any data relevant for the analysis). Additionally, references from identified publications were analysed and nonsystematic search via Google was performed. This allowed to identify another 9 studies.
Altogether, we found 16 studies reporting the % of falls among included PD patients. A brief characteristics of the studies is provided in Table 86 and data extracted from the studies -in Table 87.  The falls data reported by different studies are not consistent and there is a big difference between low and high values. In order to explain differences between reported frequency of falls, patients characteristics in each study were analysed (Table 86). One possible explanation for difference in estimated percentage of fallers can be due to duration of studies -a higher value can be expected for those with longer study duration as they can cover more subjects in later stages of PD.
The following calculations and assumptions were made to arrive at the values for model analyses: • We have calculated percentage of fallers within model cycle based on data extracted from all 16 studies, resulting in a mean value of 42.78%. As this value is similar to the value reported by Latt 2009 for UPDRS III 10-19 group -47%, it was assumed that it can be applied to patients at baseline condition in our model (average baseline UPDRS III value of 12.3). Additionally we assumed that there is only one fall per faller.
• The fraction of injurious falls was assessed based on 3 studies. Calculated mean frequency of injurious falls as a percentage of all falls is 50.9% (Table 88) Injurious falls were assumed to lead to hospitalisations. • The reviewed literature indicates that individual UPDRS I-IV scales as well as total UPDRS values are good predictors of the increase of falls frequency with PD progression [251,259,321,322].
Most data were published for UPDRS III, so this scale was chosen as a predictive factor of PD progression related falls. From the list of reviewed papers, 3 publications containing changes of frequency of falls odds ratio per unit of UPDRS III score increase with data have been selected to extract information used in our model. The averaged OR based on these studies is 1.067. A summary of final assumptions of the analysis with regard to falls is presented in the table below.

B.7. Mortality
Mortality rates in the model are based on UK Life Tables for the general population, and adjustments   for more advanced patients (defined through UPDRS III) Table 91. Additionally, for illustrative purposes, estimated percentage of males and average death probability for the cohort of patients included in the base-case analysis (age = 52.5, proportion of males = 71.3%) is shown.

B.7.2. Association between UPDRS scores and mortality
According to the results of the meta-analysis by MacLeod 2014, mortality in PD is higher than in the general population. However, studies in early PD stages indicated mortality was not different from the Based on available data, for predicting increase of mortality associated with disease progression measured by UPDRS, data on UPDRS III scores were used as this was the most frequently reported.
Based on results of 3 studies, hazard ratio per 10 points change of UPDRS III scores was calculated -1.31 (SD 0.094). As higher mortality in PD does not apply to patients at initial stages of disease, in the model the increased risk of mortality predicted by UPDRS scores was added for patients with UPDRS III score above 15 (value chosen arbitrarily).