Optimising treatment decision rules through generated effect modifiers: a precision medicine tutorial

Eva Petkova, Hyung Park, Adam Ciarleglio, R Todd Ogden, Thaddeus Tarpey, Eva Petkova, Hyung Park, Adam Ciarleglio, R Todd Ogden, Thaddeus Tarpey

Abstract

This tutorial introduces recent developments in precision medicine for estimating treatment decision rules. The objective of these developments is to advance personalised healthcare by identifying an optimal treatment option for each individual patient based on each patient's characteristics. The methods detailed in this tutorial define composite variables from the patient measures that can be viewed as 'biosignatures' for differential treatment response, which we have termed 'generated effect modifiers'. In contrast to most machine learning approaches to precision medicine, these biosignatures are derived from linear and non-linear regression models and thus have the advantage of easy visualisation and ready interpretation. The methods are illustrated using examples from randomised clinical trials.

Keywords: Treatment effect modifiers; personalised treatment assignment; single index models; treatment decision rule (TDR); value of TDR.

Conflict of interest statement

Declaration of interest: None.

Figures

Fig. 1
Fig. 1
The relationship between the derived generated effect modifier (GEM) and reading achievement outcome for ParentCorps (light green) and pre-kindergarten as usual (dark green) interventions.
Fig. 2
Fig. 2
The relationship between the derived single index z = α'x and change in depression severity for placebo (dark green curve) and the drug (light green curve) treatment.
Fig. 3
Fig. 3
Values of the treatment decision rules based on the non-linear (single-index model with multiple-links (SIMML)) and linear generated effect modifier approaches, and the two trivial treatment decisions to treat everyone with the antidepressant (Drug all) or with placebo (Placebo all) with 95% confidence intervals.

References

    1. Brotman LM, Dawson-McClure S, Calzada E, Huang KY, Kamboukos D, Palamar JJ, et al. Cluster (school) RCT of ParentCorps: impact on kindergarten academic achievement. Pediatrics 2013; 131: e1521–9.
    1. Markowitz JC, Neria Y, Lovell K, Van Meter PE, Petkova E. History of sexual trauma moderates psychotherapy outcome for posttraumatic stress disorder. Depress Anxiety 2017; 34: 692–700.
    1. Zhao Y, Zeng D, Rush AJ, Kosorok MP. Estimating individualized treatment rules using outcome weighted learning. J Am Stat Assoc 2012; 107: 1106–18.
    1. Song R, Kosorok M, Zeng D, Zhao Y, Laber EB, Yuan M. On sparse representation for optimal individualized treatment selection with penalized outcome weighted learning. Stat 2015; 4: 59–68.
    1. Laber EB, Zhao YQ. Tree-based methods for individualized treatment regimes. Biometrika 2015; 102: 501–14.
    1. Robins J, Rotnitzky A, Zhao L. Estimation of regression coefficients when some regressors are not always observed. J Am Stat Assoc 1994; 89: 846–66.
    1. Zhang B, Tsiatis AA, Laber EB, Davidian M. A robust method for estimating optimal treatment regimes. Biometrics 2012; 68: 1010–8.
    1. Gueorguieva R, Mallinckrodt C, Krystal JH. Trajectories of depression severity in duloxetine clinical trials: insights into placebo and antidepressant responses. Arch Gen Psychiatry 2011; 68: 1227–37.
    1. Smagula SF, Butler MA, Anderson SJ, Lenze EJ, Dew MA, Mulsant BH, et al. Antidepressant response trajectories and associated clinical prognostic factors among older adults. JAMA Psychiatry 2015; 72: 1021–8.
    1. Kelley ME, Dunlop BW, Nemeroff CB, Lori A, Carrillo-Roa T, Binder EB, et al. Response rate profiles for major depressive disorder: characterizing early response and longitudinal nonresponse. Depress Anxiety 2018; 35: 992–1000.
    1. Paul R, Andlauer TFM, Czamara D, Hoehn D, Lucae S, Pütz B, et al. Treatment response classes in major depressive disorder identified by model-based clustering and validated by clinical prediction models. Transl Psychiatry 2019; 9: 187.
    1. Tarpey T, Petkova E, Zhu L. Stratified psychiatry via convexity-based clustering with applications towards moderator analysis. Stat Interface 2016; 9: 255–66.
    1. Petkova E, Tarpey T, Su Z, Ogden RT. Generated Effect Modifiers (GEMs) in randomized clinical trials. Biostatistics 2017; 18: 105–18.
    1. R Core Team. R: A Language and Environment for Statistical Computing R Foundation for Statistical Computing, 2019. ().
    1. Su Z, Petkova E. pirate: Generated Effect Modifier (GEM) R package version 1.0.0, 2016. ().
    1. Green P, Silverman B. Nonparametric Regression and Generalized Linear Models: A Roughness Penalty Approach. Chapman & Hall/CRC Press, 1994.
    1. Brillinger RD. A generalized linear model with ‘Gaussian’ regressor variables In A Festschrift for Erich L. Lehman (eds Bickel PJ, Doksum KA and Hodges JL). Wadsworth, 1982.
    1. Stoker TM. Consistent estimation of scaled coefficients. Econometrica 1986; 54: 1461–81.
    1. Park HG, Petkova E, Tarpey T, Ogden RT. A single-index model with multiple-links. J Stat Plan Inference 2020; 205: 115–28.
    1. de Boor C. A Practical Guide to Splines. Springer-Verlag, 2001.
    1. Park HG, Petkova E, Tarpey T, Ogden RT. simml: Single-Index Models with Multiple-Links. R package version 0.1.0, 2019. ().
    1. Trivedi MH, McGrath P, Fava M, Parsey RV, Kurian B, Phillips ML, et al. Establishing moderators and biosignatures of antidepressant response in clinical care (EMBARC): rationale and design. J Psychiatr Res 2016; 78: 11–23.
    1. Herrera-Guzman I, Guidayol-Ferre E, Herrera-Guzman D, Guardia-Olmos J, Hinojosa-Calvo E, Herrera-Abarca JE. Effects of selective serotonin reuptake and dual serotonergic–noradrenergic reuptake treatments on memory and mental processing speed in patients with major depressive disorder. Psychiatr Res 2009; 43: 855–63.
    1. Deary IJ, Liewald D, Nissan J. A free, easy-to-use, computer-based simple and four-choice reaction time programme: the Deary-Liewald reaction time task. Behav Res Methods 2011; 43: 258–68.
    1. Loonstra A, Tarlow AR, Sellers AH. Controlled Oral Word Association Test (COWAT) metanorms across age, education, and gender. Appl Neuropsychol 2001; 8: 161–6.
    1. Flanker BA, Eriksen CW. Effects of noise letters upon identification of a target letter in a non-search task. Percept Psychophys 1974; 16: 143–9.
    1. Cai T, Tian L, Wong PH, Wei LJ. Analysis of randomized comparative clinical trial data for personalized treatment selection. Biostatistics 2011; 12: 270–82.
    1. Qian M, Murphy SA. Performance guarantees for individualized treatment rules. Ann Stat 2011; 39: 1180–210.
    1. Wang Y, Fu H, Zeng D. Learning optimal personalized treatment rules in consideration of benefit and risk: with an application to treating type 2 diabetes patients with insulin therapies. J Am Stat Assoc 2018; 113: 1–13.
    1. Benkeser D, Ju C, Lendle S, van der Laan M. Online cross-validation-based ensamble learning. Stat Med 2018; 37: 249–60.
    1. Ramsay JO, Silverman BW. Functional Data Analysis (2nd edn). Springer, 2005.
    1. Ciarleglio A, Petkova E, Tarpey T, Ogden RT. Treatment decisions based on scalar and functional baseline covariates. Biometrics 2015; 71: 884–94.
    1. Ciarleglio A, Petkova E, Tarpey T, Ogden RT. Flexible functional regression methods for estimating individualized treatment regimes. Stat (Int Stat Inst) 2016; 5: 185–99.
    1. Ciarleglio A, Petkova E, Tarpey T, Ogden RT. Constructing treatment decision rules based on scalar and functional predictors when moderators of treatment effect are unknown. J R Stat Soc Ser C 2018; 67: 1331–56.
    1. Laber EB, Staicu AM. Functional feature construction for individualized treatment regimes. J Am Stat Assoc 2018; 113: 1219–27.

Source: PubMed

Szponzorok és közreműködők

Egészségi állapot

Kábítószer-beavatkozások

3
Iratkozz fel