PROMIS Physical Function: Item Response Theory (IRT) CAT, T-Score Interpretation (Mean 50, SD 10), Global-10, and HL7 FHIR Integration

PROMIS Physical Function is the foundational physical health domain within the National Institutes of Health (NIH) Patient-Reported Outcomes Measurement Information System (PROMIS)—a standardized psychometric initiative launched under the NIH Roadmap for Medical Research (Cella et al., 2007; Rose et al., 2008) and stewarded by Northwestern University’s HealthMeasures. Unlike legacy disease-specific questionnaires that force every patient to answer the same fixed list of items, the adult PROMIS Physical Function item bank (v2.0) calibrates more than 121 five-point Likert items across four functional subdomains—Mobility (Lower Extremity), Upper Extremity (Dexterity), Axial/Central (Neck and Back), and Instrumental Activities of Daily Living (IADL)—onto a single common metric.

Across orthopaedic surgery, rheumatology, oncology, and ePRO in clinical trials, clinical investigators can administer PROMIS Physical Function as a 4- to 12-item Computerized Adaptive Test (CAT), a fixed short form (4a, 6b, 8b, 10a, or 20a), or part of multi-domain profiles such as PROMIS-29 and PROMIS Global-10, with all administration modes producing directly comparable T-scores (mean = 50, standard deviation = 10).

How Does Item Response Theory (IRT) Computerized Adaptive Testing (CAT) Work in NIH PROMIS Instruments?

Whereas Classical Test Theory (CTT) instruments (such as the WOMAC or KOOS) sum ordinal raw scores and suffer from floor effects in severely disabled inpatients and ceiling effects in high-functioning athletes, NIH PROMIS item banks are calibrated using Item Response Theory (IRT)—specifically Samejima’s (1969) two-parameter Graded Response Model (GRM).

Under the Graded Response Model, every item in the PROMIS Physical Function bank is characterized by two calibrated parameter sets along the latent trait continuum (θ, theta):

• Item Discrimination (Slope, a): Quantifies how sharply the item differentiates between respondents with slightly different levels of physical capability.
• Category Thresholds (Location / Difficulty, b₁, b₂, b₃, b₄): Identify the exact points along the physical function continuum (from bedbound inability to turn in bed up to running 5 miles or performing strenuous sports) where a respondent has a 50% probability of endorsing a higher response category on the 5-point scale (5 = Without any difficulty, 4 = With a little difficulty, 3 = With some difficulty, 2 = With much difficulty, 1 = Unable to do).

When delivered as a Computerized Adaptive Test (CAT) via an electronic clinical outcome assessment (eCOA) engine or EHR patient portal:

1. Starting Item: The CAT algorithm presents an initial item calibrated near the U.S. general population mean (θ = 0, T-score = 50), such as walking at a normal speed or climbing a flight of stairs.
2. Bayesian Posterior Updating (EAP) & Maximum Fisher Information: After each answer, the engine calculates an interim Expected A Posteriori (EAP) estimate of the patient’s latent physical function (θ) and its standard error (SE), then dynamically selects the next unadministered item from the 121-item bank that maximizes Fisher information at that exact θ estimate.
3. Standard CAT Stopping Rule (4 to 12 Items): Under standard HealthMeasures PROMIS CAT parameters, administration stops automatically as soon as at least 4 items have been answered AND the Standard Error (SE) on the theta metric falls below 0.30 (equivalent to SE < 3.0 on the T-score metric, corresponding to a reliability coefficient ≥ 0.91), or when a maximum of 12 items has been reached.

How Do You Interpret a PROMIS T-Score (General Population Mean = 50, SD = 10) for Physical Function vs. Pain Interference and Depression?

All PROMIS instruments convert the IRT latent trait estimate (θ, which has a mean of 0 and SD of 1) into a standardized T-score using the linear transformation T = 50 + (10 × θ), centered on the 2000 U.S. Census general population norm (where Mean = 50 and Standard Deviation [SD] = 10).

A critical clinical rule in PROMIS interpretation is T-score directionality—a higher PROMIS T-score always represents more of the concept being measured:

• Function Domains (Higher T-Score = Better Health): For PROMIS Physical Function, Upper Extremity, Mobility, Cognitive Function, and Ability to Participate in Social Roles and Activities, a T-score above 50 indicates better function than the U.S. population average, whereas a T-score below 50 indicates functional impairment.
• Symptom Domains (Higher T-Score = Worse Health): For PROMIS Pain Interference, Fatigue, Depression, Anxiety, and Sleep Disturbance, a T-score above 50 indicates worse symptom burden than the U.S. population average, whereas a T-score below 50 indicates fewer symptoms.

Severity Category	SD Distance from Population Mean (50)	PROMIS Physical Function T-Score Cutoff (Higher = Better Function)	PROMIS Pain Interference, Fatigue, Depression & Anxiety T-Score Cutoff (Higher = Worse Symptoms)
Within Normal Limits	Better than -0.5 SD (Function) / < +0.5 SD (Symptoms)	≥ 45.0 (50 = U.S. Mean; 60 = +1 SD above average)	< 55.0 (50 = U.S. Mean; 40 = -1 SD fewer symptoms)
Mild Impairment / Mild Symptoms	0.5 to 1.0 SD worse than mean	40.0 to 44.9	55.0 to 59.9
Moderate Impairment / Moderate Symptoms	1.0 to 2.0 SD worse than mean	30.0 to 39.9	60.0 to 69.9
Severe Impairment / Severe Symptoms	> 2.0 SD worse than mean	< 30.0	≥ 70.0

Across orthopedic and rheumatologic cohorts, the Minimal Clinically Important Difference (MCID) for PROMIS Physical Function typically ranges from 3.0 to 5.0 T-score points (0.3 to 0.5 SD).

How Are PROMIS-29 and PROMIS Global-10 (Hays et al.) Mapped to EQ-5D Utilities and Integrated via HL7 FHIR Questionnaire?

When healthcare systems require a rapid global snapshot of health-related quality of life (HRQoL) across all clinical service lines without administering domain-specific CATs, they deploy the 10-item PROMIS Scale v1.2 – Global Health (PROMIS Global-10), developed by Ron D. Hays and colleagues (2009):

• Global Physical Health (GPH) Summary T-Score (4 Items): Computed from IRT calibration of Global03 (physical health rating, 1–5), Global06 (ability to carry out everyday physical activities, 1–5), Global07r (0–10 numerical pain rating recoded to 5 categories), and Global08r (fatigue severity recoded 1–5).
• Global Mental Health (GMH) Summary T-Score (4 Items): Computed from Global02 (quality of life, 1–5), Global04 (mental health including mood and ability to think, 1–5), Global05 (satisfaction with social activities and relationships, 1–5), and Global10r (frequency of emotional problems such as anxiety, depression, or irritability).
• Single-Item General Health & Social Roles (2 Unscored Items): Global01 (In general, would you say your health is…) and Global09r (ability to carry out usual social activities and roles) are retained for individual tracking and cross-study comparison with the SF-36.
• Mapping to EQ-5D Health Utilities & QALYs: For cost-effectiveness and quality-adjusted life-year (QALY) health economics modeling, validated regression algorithms—including Revicki et al. (2009) for PROMIS Global-10 and the PROPr (PROMIS-Preference) multi-attribute utility scoring system (Dewitt et al., 2018) for PROMIS-29—map PROMIS domain T-scores directly onto EQ-5D-3L / EQ-5D-5L index utilities and standard gamble/time trade-off weights.

HL7 FHIR Questionnaire, SDC Adaptive Questionnaire, and LOINC Observation Mapping

In modern health IT architectures using the HL7 FHIR Structured Data Capture (SDC) Implementation Guide and HL7 FHIR Observation profiles:

• Fixed Forms vs. Adaptive CATs: Fixed short forms (such as PROMIS Physical Function 10a or PROMIS Global-10) are represented as standard FHIR Questionnaire definitions and captured as QuestionnaireResponse resources. Dynamic CAT assessments use the FHIR SDC Adaptive Questionnaire ($next-question operation), where the EHR client sends each answered item in a QuestionnaireResponse to the HealthMeasures Assessment Center API and receives the next dynamically selected question until the SE < 0.30 stopping criterion is met.
• LOINC-Coded FHIR Observation Output: Once scored, the resulting PROMIS Physical Function T-score (e.g., LOINC panel/score codes under the PROMIS hierarchy, such as LOINC 91721-1 for PROMIS Physical Function T-score or LOINC 71969-0 / 71967-4 for Global Physical and Mental Health T-scores) is written to the EHR as a discrete Observation (`category = survey`, `valueQuantity.value = T-score`, `unit = “{score}”`).

Frequently Asked Questions

How does Item Response Theory (IRT) Computerized Adaptive Testing (CAT) work in NIH PROMIS instruments?

NIH PROMIS instruments calibrate item banks using Samejima’s Graded Response Model (an Item Response Theory model). During a Computerized Adaptive Test (CAT), the algorithm starts with an item near the population average (T-score = 50) and, after each patient response, updates the latent trait estimate (theta) and dynamically selects the next item that provides maximum Fisher information at that ability level. The CAT automatically stops once at least 4 items (and at most 12 items) are answered and the standard error drops below 3.0 T-score points (SE < 0.30 on the theta scale).

How do you interpret a PROMIS T-score (general population mean = 50, SD = 10) for Physical Function vs. Pain Interference and Depression?

All PROMIS T-scores are standardized to a U.S. general population mean of 50 and standard deviation (SD) of 10, where a higher score always means more of the measured concept. Therefore, for PROMIS Physical Function, a T-score above 50 is better than average, while 40–44.9 indicates mild impairment, 30–39.9 moderate impairment, and below 30 severe impairment. Conversely, for symptom domains like Pain Interference and Depression, a T-score above 50 is worse than average (55–59.9 mild, 60–69.9 moderate, and ≥70 severe symptoms).

How are PROMIS-29 and PROMIS Global-10 (Hays et al.) mapped to EQ-5D utilities and integrated via HL7 FHIR Questionnaire?

PROMIS Global-10 (Hays et al., 2009) generates Global Physical Health (GPH) and Global Mental Health (GMH) T-scores from two 4-item subsets, which—along with PROMIS-29—can be converted into EQ-5D health utility weights and QALYs via the Revicki et al. linear mapping or the PROPr utility algorithm. Technically, EHRs administer fixed forms via HL7 FHIR Questionnaire/QuestionnaireResponse or CATs via the FHIR SDC $next-question adaptive operation, storing the final LOINC-coded T-score as a discrete FHIR Observation.