The impact of treatment line matching on covariates balance and cost effectiveness results: A case study in oncology

Size: px

Start display at page:

Download "The impact of treatment line matching on covariates balance and cost effectiveness results: A case study in oncology"

Kimberly Sanders
5 years ago
Views:

1 The impact of treatment line matching on covariates balance and cost effectiveness results: A case study in oncology Xavier G.L.V. Pouwels Bram L. T. Ramaekers Manuela A. Joore Department of Clinical Epidemiology and Medical Technology Assessment Maastricht University Medical Centre+ The case (1/2) Population: Metastatic breast cancer patients (after 2 or more previous chemotherapies) Intervention: Treatment X Comparator: Usual care Outcome: Costs per quality-adjusted life years (QALYs) Main data source: clinical practice data No randomisation Confounding by indication

2 The case (2/2) Control Diagnosis Tx2 A Tx2 A Eligible for X Tx3 B Tx4 C Treatment line matching = using different version of the Eligible for X same patient Tx3 ( matching Tx4 Tx5 with X B C replacement). SELECTION Patients not similar Effectiveness of X may vary Intervention Tx3 B Tx3 B Tx4 X Tx4 C Tx5 C Advantage: increased number of potential matches. Tx5 X 3 Research question When using propensity score matching or genetic matching, what is the influence of using different treatment lines instead of patients on the covariates balance and ultimately the cost effectiveness of treatment X versus?

3 Methods Matching procedure Matching Levels Unmatched Genetic Matching Propensity score Matching Not applicable X X Patient level matching Treatment line level matching X X Matching 1:1 with replacement ( Matching R package) Methods - Matching procedure Treatment X VS Unmatched N=251 Individual patients N=251 Individual treatment lines N=440 Patient GenMatch Patient PSMatch Treatment line GenMatch Treatment line PSMatch

4 Methods - Matching procedure Covariate balance assessed based on median, variance ratio, eqq plots, and bootstrapped Kolmogorov- Smirnov tests. Treatment X VS Unmatched N=251 Individual patients N=251 Individual treatment lines N=440 Patient GenMatch Patient PSMatch Treatment line GenMatch Treatment line PSMatch Methods Treatment effectiveness input Parametric time-to-event models Progression-free survival (PFS) Overall survival (OS) 7 distributions, selection for base-case based on NICE DSU TSD 14. Same distribution for PFS and OS in all groups.

5 Methods Cost effectiveness analysis Health states: Progression-free, progressed, dead Perspective: Dutch Health care Cycle length: 1 week Utility values: Literature Health care resources: Treatments, outpatients, hospitalisations Outcomes: Incremental cost effectiveness ratio Incremental net monetary benefits Results Covariate balance comparison Treatment X Unmatched GenMatche d patient GenMatche d treatment line PSMatched PSMatched patient treatment line Number of patients Number of previous hormonal therapy % % 26 43% 27 45% 21 35% 27 45% % 90 36% 21 35% 19 32% 24 40% 20 33% % 20 8% 13 22% 14 23% 15 25% 13 22% Median KS bootstrap In general, GenMatch performed better than p-value N.A Number of previous PSMatch, chemotherapy and covariates balance were improved in 0-2 the treatment 20 33% 221 line 88% matched 24 40% groups 18 30% compared 31 52% to 29 48% the 50% patient 24 10% level 28 matched 47% 35 groups. 58% 19 32% 24 40% % 6 2% 8 13% 7 12% 10 17% 7 12% Median KS bootstrap p-value N.A

6 Results Matching (start age, GenMatch) Unmatched Patient Treatment line Results Effectiveness input (Gompertz distribution, OS)

Results Resource use and costs inputs Total weekly costs PFS PD Unmatched 1,162 719 Patient GenMatched 808 196 Treatment line GenMatched 781 475 Patient PSMatched 1,186 222 Treatment line PSMatched

7 Results Resource use and costs inputs Total weekly costs PFS PD Unmatched 1, Patient GenMatched Treatment line GenMatched Patient PSMatched 1, Treatment line PSMatched 1, Results costs effectiveness (probabilistic) Life years Quality adjusted life years Costs Treatment X ,019 Unmatched ,005 GenMatched Patient ,462 GenMatched Treatment line usual 19,755 care PSMatched Patient ,320 PSMatched Treatment line usual care ,697 Incremental Comparison LY QALY Costs ICER QALY inmb (WTP = 80,000) Intervention versus unmatched ,985 67,908 3,688 Intervention versus Patient GenMatched , ,175-6,149 Intervention versus Treatment line GenMatched ,265 4,525,276-13,030 Intervention versus Patient PSMatched ,699 Dominated - 12,347 Intervention versus Treatment line PSMatched ,678 44,990-4,418

8 Results Cost effectiveness plane Results Cost effectiveness plane

9 Results Cost effectiveness plane Results Cost effectiveness acceptability curves

10 Conclusions (1/2) Unmatched Patient PSMatched Treatment-line PS Matched Patient GenMatched Treatment-line GenMatched Matching procedures can have substantial impact on model inputs and results. Uncertainty Decision making Conclusions (2/2) Generalisability? Treatment line matching seems to be a viable option to increase the number of potential matches when the number of patients in the comparator group is small.

11 Thank you for your attention! Acknowledgement: E. Koffijberg Statements (Richard Grieve) Non-randomised studies should consider several approaches as part of structural sensitivity analyses.

12 Statements (Wietske Kievit) We should rely more on observational comparisons in guideline development and health technology assessment. Statements (Xavier Pouwels) Treatment line matching should always be considered when nonrandomised comparative evidence is used to inform cost effectiveness analysis.

13 Statements (Richard Grieve) Genetic matching is an attractive approach for balancing observed confounders. Statements (Wietske Kievit) Residual confounding may be substantial even when good prognostic data are available and is an RCT always necessary for comparative effectiveness questions.

14 Statements (Xavier Pouwels) Since comparisons to an unmatched group are biased, as much as a complete case analysis could be, these analyses are uninformative and should not be performed. Statements (Richard Grieve) Move to large e-health data offers opportunity for IV methods that fully recognise heterogeneity.

Background 1. Comparative effectiveness of nintedanib

Background 1. Comparative effectiveness of nintedanib NCPE report on the cost effectiveness of nintedanib (Vargatef ) in combination with docetaxel for the treatment of adult patients with locally advanced, metastatic or locally recurrent non-small cell lung