Comparative Validity of the American Speech-Language-Hearing Association's National Outcomes Measurement System,...

Background The American Speech-Language-Hearing Association (ASHA) advocates using the National Outcomes Measurement System (NOMS) scales to assist speech-language pathologists (SLPs) in the mandated assigning of "functional limitation levels" and G-Code for Medicare Part B claims. Unfortunately, not all SLPs have access to this tool, and it is unclear how other common outcome measurement scales relate to ASHA NOMS or G-Codes. To explore the utility of other scales in reporting Centers for Medicare & Medicaid Services G-Codes, we conducted a comparative validity study comparing ASHA NOMS Dysphagia Scale, Functional Oral Intake Scale (FOIS), and Mann Assessment of Swallowing Ability to G-Codes on a sample of 105 patients who had stroke. Method Nine SLP student researchers trained to criterion on the NOMS and FOIS blindly and independently scored 105 stroke cases with dysphagia de-identified from a past study. Three graduate SLP clinicians independently assigned G-Codes. Data from the student researchers and Mann Assessment of Swallowing Ability scores were compared for admission and discharge from subacute rehabilitation. Analysis included intraclass correlation for reliability, Spearman correlation for comparative validity, and area under the receiver operating characteristic curve to compare the validity and discriminatory utility of measures in classifying dysphagia. Results Strong correlations (> .6) were noted between all measures at baseline, particularly FOIS and NOMS coding ( r = .919). At discharge, superior performance by the FOIS (area under the receiver operating characteristic curve = 0.819) was demonstrated. Code missingness was higher for the NOMS than the other scales. Discussion All 3 clinical dysphagia tools demonstrate acceptable validity in supporting G-Code designation to stroke cases. The FOIS demonstrated superior validity and utility across time points. The NOMS Dysphagia Scale was significantly affected by data missingness due to the multiconstruct nature of the tool.