CHAPTER 1 Geriatric Physical Therapy in the 21st Century 11 providing plausible alternative explanations for the ob defined)an odds ratio greater than 3 is d a There are several distinguishing features of quality in up:an odds ratio greater than 10 as a verv a systematic review.A systematic review should confirm large increase.Odds ratios less than 1(identified as that a comprehen ve search of the appropriate literature negative odd ratios)that the prs the been pe repro included studies mee t established inclusion criteria.At negative odds ios is 1 to 0.An odds ratio of 0.7 is least two reviewers should independently assess quality generally described as representing a moderate decrease med ).2 as a ver arge fde in od and odds rati ng ih th rified.The ndations nd stat of the ted as the ide strength of the evidence are well grounded and clearly In order for an odds ratio to be considered statisticall nchddonithegais,imdngandappeb (and thus generalizable the scores withi rtance of the Findings as as core 11。 of the study detailed discussion of statistical Diagnosis Studies.Sensitivity,specificity,and likeli. studies is found elsewhere 20 comn nonly reporte hndings of In comparison to logistic regression, aim est 1n6 out ome ong a tha is high a ent location within one of tw likely to rule out the condition,whereas,when specificit ontinuum of s cores based on scores tion 6 bile best f o6eaCrosamcea on pre Fo age or negative test effect with ha onditi edict the gait d of co delling older adults.The outcom of linear r hood ratio (LR+)( rbitrarily identified as a score above would be an equation that can be used to predict the 10)in svery likely to be specific gait or omparabl given the rily id of by the del indicate the des a score below 0.2)indicates that it is very unlikely that to which all the variables included in the model acc the person with a negative test has the condition. Prognosis Studies Progno studie examine the scor cted ely,the ex al tical analysis of choice is a regression analysis.Logistic provides useful information about trends in the popula- regression is utilized more commonly than linear regres- sion because many of the one specific patient. ce or st may ell great in smal f variables.The aim of progn c studies using istic significant predictions that regression is account for as little as 40%of the variance may have some a set pred ments alue in guiding ju rela for 0 of the variance would be perceived as very compelling the end of rehabilitation (as compared to those who go hndings. ns T he mo variability in the predic tor va abl ly the ase in of old ad he is is thu ables are examined and.in combination. ovide a sta. tistically more robust assessment of the odds of obtain- dence in the accuracy of the prediction.Studies may need particularly large sample sizes combin with a large number of w osen predictor variables to explair CHAPTER 1 Geriatric Physical Therapy in the 21st Century 11 providing plausible alternative explanations for the ob￾served outcomes. There are several distinguishing features of quality in a systematic review. A systematic review should confirm that a comprehensive search of the appropriate literature has been performed using a transparent and reproduc￾ible process for identifying studies and confirming that included studies meet established inclusion criteria. At least two reviewers should independently assess quality and applicability of each study considered for the review. Meta-analysis across studies is performed if sufficient numbers of studies with sufficient homogeneity are iden￾tified. The recommendations and statement of the strength of the evidence are well grounded and clearly justified based on the quality, findings, and applicability of the included studies. Determining the Importance of the Findings of the Study Diagnosis Studies. Sensitivity, specificity, and likeli￾hood ratios are the most commonly reported findings of studies aimed at establishing the accuracy of diagnostic tools. Several references provide excellent reviews of this topic.2,17 When sensitivity is high, a negative test result is likely to rule out the condition, whereas, when specificity is high, a positive test result is likely to rule in the condi￾tion. Likelihood ratios (LRs) are best for increasing the therapist’s confidence in the ability to associate a positive or negative test effect with having the target condition/ disorder (posttest probability).20 A high positive likeli￾hood ratio (LR1) (arbitrarily identified as a score above 7 or 10) indicates that the condition is very likely to be present in the person with a positive test. Conversely, a very low likelihood ratio (LR2) (arbitrarily identified as a score below 0.2) indicates that it is very unlikely that the person with a negative test has the condition. Prognosis Studies. Prognosis studies examine the ability of selected factors to predict an outcome of inter￾est. Most commonly, although not exclusively, the statis￾tical analysis of choice is a regression analysis. Logistic regression is utilized more commonly than linear regres￾sion because many of the key explanatory variables (e.g., “sex” or “presence or absence of surgical history”) as well as the outcome of interest are categorical variables. The aim of prognostic studies using logistic regression is to determine the extent to which the presence or absence of selected variables predicts a pa￾tient’s outcome or risk of belonging to a target group. For example, how accurately does a set of prognostic variables predict which subjects are likely to go home at the end of rehabilitation (as compared to those who go to a nursing home or other setting)? These predictions provide an estimate of the “odds” of belonging in the target outcome group. Typically, several predictor vari￾ables are examined and, in combination, provide a sta￾tistically more robust assessment of the odds of obtain￾ing an outcome (i.e., belong to the target group) than one variable alone. By convention (and fairly arbitrarily defined) an odds ratio greater than 3 is generally inter￾preted as a moderate increase in odds of being in the target group; an odds ratio greater than 10 as a very large increase. Odds ratios less than 1 (identified as negative odds ratios) indicate that the presence of the predictor variables is related to decreased odds of being in the target group. The full range of possible scores for negative odds ratios is 1 to 0. An odds ratio of 0.7 is generally described as representing a moderate decrease in odds of being in the target group, and an odds ratio of 0.2 as a very large decrease in odds of being in the target group. The confidence interval (CI), most com￾monly reported as the 95% CI, must also be considered. In order for an odds ratio to be considered statistically significant (and thus generalizable), the scores within the bracketed CI must NOT include 1, as a score of 1 represents equal odds of being in either group. A more detailed discussion of statistical analysis and prognosis studies is found elsewhere.20 In comparison to logistic regression, linear regression examines outcomes along a continuum. Rather than focusing on whether or not a set of variables can predict patient location within one of two identified groups, a linear regression analysis wants to determine a specific score across a linear continuum of scores based on scores on predictor variables. For example, patient age, heart rate, and number of chronic health conditions might be hypothesized to predict the gait speed of community￾dwelling older adults. The outcome of linear regression would be an equation that can be used to predict the specific gait speed of comparable patients given their scores on each of the predictor variables. The proportion of variance explained by the model indicates the degree to which all the variables included in the model account for the outcome or dependent variable. A model that predicts the outcome score perfectly would be described as explaining all the variance; however, realistically, there is always unexplained variance. Linear regression provides useful information about trends in the popula￾tion but is often not very useful in predicting the scores of one specific patient. Variability among and between subjects may be too great in small, convenience samples, which is typically the case in the rehabilitation literature. Generally, statistically significant predictions that account for as little as 40% of the variance may have some value in guiding judgments about the relative contributions of a set of predictor variables, and a study that constructed a predictive model accounting for 70% of the variance would be perceived as very compelling findings. The more variability in the predictor variables—as is commonly the case in studies of older adults—the less robust the prediction, thus lowering the odds ratio or percentage of variance explained, which decreases confi￾dence in the accuracy of the prediction. Studies may need particularly large sample sizes combined with a large number of well-chosen predictor variables to explain