Propensity score

Later Sweredoski et al. (37) incorporated a combination of amino acid propensity scores and half-sphere exposure values at multiple distances to form the BEpro tool (formerly called PEPITO). Using the Epitopia algorithm, Rubinstein et al. (38) for the first time truly exploited an extensive set of physicochemical and structural geometrical features from an antigen s primary or tertiary structures. They trained the Naive Bayes classifier using a benchmark dataset of 66 and 194 validated nonredundant epitopes derived from antibody-antigen structures and antigen sequences,... 

BEpro (formerly known as PEPITO) Combination of amino acid propensity scores and half-sphere exposure http / / pepito. proteomics. ics. uci.edu/ Sweredoski and Baldi (37)... 

Swanson, J., Hinshaw, S., Arnold, L., Gibbons, R., Marcus, S., Hur, K., Jensen, R, Vitiello, B., Abikoff, H., Greenhill, L., Hechtman, L., Pelham, W., Wells, K., Conners, C., March, J., Elliott, G., Epstein, J., Hoagwood, K., Hoza, B., Molina, B., Newcorn, J., Severe, J., 6c Wigal, T. (2007b). Second evaluation of MTA 36-month outcomes Propensity score and growth mixture model analyses. Journal of the American Academy of Child Adolescent Psychiatry 46, 989-1002. 

Milton RC, Sperduto RD, Clemons TE, et al. Age-Related Eye Disease Smdy Research Group. Centrum use and progression of age-related cataract in the Age-Related Eye Disease Smdy a propensity score approach. AREDS Report No. 21. Ophthalmology 2006 113 1264-1270. 

A very useful approach to augmenting informative dropout or truncated data considered here is the propensity-adjusted multiple imputation approach (25). This approach utilizes the method of reducing a multivariate stratification to a univariate stratification using the propensity score (26, 27). The propensity score is the conditional probability of assignment to a particular treatment given a vector of observed covariates. That is, at time t a subject s propensity score is defined as the probability of the subject to remain in the study through time t given the subject s... 

P. R. Rosenbaum and D. B. Rubin, The central role of the propensity score in observational studies for causal effect. Biometrika 70 41-45 (1983). 

The propensity score is a superior alternative to adjusting for confounders than analysis of covariance... 

In an influential paper in Biometrika, Rosenbaum and Rubin discussed the possibility of using what they called the propensity score for adjusting for confounders (Rosenbaum and Rubin, 1983). The idea is that a stratification be made on the probability of assignment to treatment or control as a function of covariates. In a classic randomized clinical trial, of course, the probability of assignment to either group is 1 /2 irrespective of any covariate. The philosophy of the propensity score would then be that there is but a single stratum, that with score 1/2, and so no adjustment is necessary. This is, of course, equivalent to the familiar argument that for a classic randomized trial no adjustment is necessary. 

In the bizarre example of the previous section, allocation to verum occurs with probability 1/4 if the baseline DBF is 95mmHg and 3/4 if DBF is lOOmmHg and so the propensity score takes on these two values. Thus, the philosophy of stratifying by the propensity score agrees perfectly with the argument we propounded there that adjustment using these strata (or equivalently using analysis of covariance) was indicated. 

Thus, I do not accept that the propensity score is a useful alternative to analysis of covariance. In my opinion it will produce either similar or inferior inferences. 

It is appropriate for me to repeat a general warning about the book that Carl-Fredrik Bur man has drawn to my attention. A number of the section headings contain statements of position. For example, Chapter 7 has a section, The propensity score is a superior alternative to adjusting for confounders than analysis of covariance . You will get a very misleading impression of the message of the book if you take these as being my position. This is a book about issues and where such statements are made it is nearly always because, as is the case with this one, 1 wish to take issue with them. 

Adjustment methods for confoimding fall into several categories stratification, mulhvariable models (such as logistic regression), and more advanced methods such as use of propensity scores or instrumental variables. 

Once variables are identified, one can confrol for them through direct matching or stratification, whereby fhe fargef and comparator groups are logically divided by the attributes of fhe covariafes. However, in a multivariable context, the data may be too sparse to provide adequate sample size to allow matching on all covariates or to provide subpopulations within each covariate-defined stratum (i.e., there may be empty cells defined by combinations of covariates). A popular tool to overcome this limitation is propensity score analysis (Rosenbaum 2002 Rubin 1997). 

As with other approaches, the propensity score model is only as good as the covariates selected to provide the adjustment. A propensity score is a single metric that is intended to account for all of the explanatory variables that predict who will receive treatment. Propensity scores generally balance observed confounders but do not necessarily produce balance on factors not incorporated into the model. Such imbalances represent a particular problem... 

See color insert.) Database heterogeneity. ARF, acute renal failure ALI, acute liver injury UGIB, upper gastrointestinal bleeding AMI, acute myocardial infarction HOPS, high-dimensional propensity score-adjusted cohort design. 

Rubin DB. 1997. Estimating causal effects from large data sets using propensity scores. Ann. Intern. Med. 127 757-763. 

Schneeweiss S, Rassen JA, Glynn RJ, Avom J, Mogim H, Brookhart MA. 2009. Highdimensional propensity score adjustment in studies of treatment effects using health care claims data. Epidemiology 20 512-522. 

Seeger JD, Kurth T, Walker AM. 2007. Use of propensity score technique to account for exposure-related covariates An example and lesson. Med. Care 45 5143-5148. 

In each of these regressions, one respects the fact that if the past tells us that the person already died, then the regression is degenerate and equals 1 (i.e., death) as well, and the update only modifies the regression conditional on not having died yet. Note that the intervention mechanism factorizes in a censoring mechanism and treatment mechanism (i.e., propensity score), and these should be fit separately. 

Observational studies In 741 patients with coronary bypass and/or valve operations transfusion of more than 5.5 units of blood products was associated with a higher incidence of re-operation, neurological, infectious, and cardiac complications, multisystem organ failure, and deaths [10 ]. However, these results were distorted by confounding by indication, because patients who receive more than 5.5 units of blood are more severely ill than those who receive fewer transfusions or none at all. Studies of the potential harmful effects of blood transfusions should be carefully corrected for disease severity, for example by propensity score matching analyses [ll ]. Among propensity-matched patients undergoing... 

The association between the use of abacavir and an increased risk of myocardial infarction has been confirmed in a prospective nationwide cohort study that included 2952 Danish HIV-infected patients taking highly active antiretroviral therapy (HAART) from 1995 to 2005 [87 =]. Hospitalization rates for myocardial infarction were 2.4 per 1000 person-years (95% Cl = 1.7, 3.4) for abacavir non-users and 5.7 per 1000 person-years (95% Cl = 4.1, 7.9) for abacavir users. The risk of myocardial infarction increased after abacavir was started (unadjusted IRR = 2.22 95% Cl = 1.31, 3.76 IRR adjusted for confounders = 2.00 95% Cl = 1.10, 3.64 IRR adjusted for propensity score = 2.00 95% Cl = 1.07, 3.76). This effect was also observed among patients who started to take abacavir within 2 years after the start of HAART and among patients who started to take abacavir as part of a triple nucleoside reverse transcriptase inhibitor (NRTI) regimen. 

Infection risk In a retrospective cohort study of 1677 patients undergoing coronary artery bypass surgery, in which preoperative aspirin -I- clopidogrel was compared with aspirin alone, clopidogrel was associated with an increased risk of postoperative surgical site infection and bacteremia, both unadjusted (HR = 1.51 95% Cl = 1.09, 2.08) and after adjustment for demographic, socioeconomic, preoperative, and intraoperative risk factors (HR = 1.42 95% Cl = 1.01, 2.00) and propensity score (HR = 1.43 95% Cl = 1.01, 2.01) [166< ]. 

Ngaage DL, Cale AR, Cowen ME, Griffin S, Guvendik L. Aprotinin in primary cardiac surgery operative outcome of propensity score-matched study. Ann Thorac Surg 2008 86(4) 1195-202. 

All methods used to account for households behavioral responses to transfers are data intensive. Most of the CCT evaluations from World Bank (forthcoming) are randomized evaluations, with information collected from treatment and control groups. Jalan and Ravallion (2003) use propensity score matching to evaluate the distributional outcomes of Argentina s Trabajar program. Ravallion, van de Walle, and Gautam (1995) and van de Walle (2003) use panel data and instrumental variable models to estimate a reduced form equation of household consumption on transfer incomes. 

Estimating the Benefit Incidence of an Antipoverty Program by Propensity-Score... 

Cheng SY, et al Evaluation of inpatient multidisciplinary palliative care unit on terminally ill cancer patients from providers perspectives a propensity score analysis. Jpn J Clin Oncol 2013 43 161 169. 

A review of 20 clinical trials published between 1979 and 2012 showed that subjects randomised to treatment with mixed amphetamine salts demonstrated a statistically significant increase in the resting heart rate [+5.7 bpm (3.6, 7.8), p < 0.0001] and systolic blood pressure [+2.0 mmHg (0.8, 3.2), p = 0.005] compared to subjects randomised to placebo [1 ]. A significant increased risk in resting heart rate >90 bpm [4.2% (n = 50) versus 1.7% (n = 8), odds ratio (OR) = 2.75 (1.3,6.7), p = 0.006] was found [l ]. A retrospective cohort study of new amphetamine users (n = 38,586) showed that the propensity score-adjusted hazard ratio for sudden death/ventricular arrhythmia was 1.18 (95% confidence interval [Cl] 0.55-2.54), for stroke 0.80 (95% Cl 0.44-1.47), for myocardial infarction 0.75 (95% Cl 0.42-1.35), and 0.78 for stroke/myocardial infarction (95% CEO.51-1.19) [2 ]. In a review of claims of privately insured young people, ages 6-21 years, without known cardiovascular risk factors (n = 171,126) that included all methylphenidate and... 

A retrospective cohort study of new atomoxetine users (n = 20,995) showed that the propensity-score adjusted hazard ratio for sudden death/ventricular arrhythmia was 0.41 (95% Cl 0.10-1.75), for stroke 1.30 (95% Cl 0.52-3.29), for myocardial infarction 0.56 (95% CLO.16-2.00) and 0.92 for stroke/myocardial 0.75 infarction (95% CEO.44-1.92) [2 ]. For atomoxetine, the most commonly reported ADRs in >5% of patients using the agent were decreased appetite, weight loss, suicidal thoughts in children, abdominal pain, erectile dysfunction, headache, insomnia, nausea, xerostomia and tachycardia [6 ] [7 j [36 j [37 ]. 

Chatteq ee S, Chen H, Johnson ML, Aparasu RR. Comparative risk of cerebrovascular adverse events in community-dwelling older adults using risperidone, olanzapine and quetiapine a multiple propensity score-adjusted retrospective cohort study. Drugs Aging... 

---