Advertisement

Mayo-Portland Adaptability Inventory: Comparing Psychometrics in Cerebrovascular Accident to Traumatic Brain Injury

      Abstract

      Malec JF, Kean J, Altman IM, Swick S. Mayo-Portland Adaptability Inventory: comparing psychometrics in cerebrovascular accident to traumatic brain injury.

      Objectives

      (1) To evaluate the measurement reliability and construct validity of the Mayo-Portland Adaptability Inventory, 4th revision (MPAI-4) in a sample consisting exclusively of patients with cerebrovascular accident (CVA) using single parameter (Rasch) item-response methods; (2) to examine the differential item functioning (DIF) by sex within the CVA population; and (3) to examine DIF and differential test functioning (DTF) across traumatic brain injury (TBI) and CVA samples.

      Design

      Retrospective psychometric analysis of rating scale data.

      Setting

      Home- and community-based brain injury rehabilitation program.

      Participants

      Individuals post-CVA (n=861) and individuals with TBI (n=603).

      Interventions

      Not applicable.

      Main Outcome Measure

      MPAI-4.

      Results

      Item data on admission to community-based rehabilitation were submitted to Rasch, DIF, and DTF analyses. The final calibration in the CVA sample revealed satisfactory reliability/separation for persons (.91/3.16) and items (1.00/23.64). DIF showed that items for pain, anger, audition, and memory were associated with higher levels of disability for CVA than TBI patients; whereas, self-care, mobility, and use of hands indicated greater overall disability for TBI patients. DTF analyses showed a high degree of association between the 2 sets of items (R=.92; R2=.85) and, at most, a 3.7 point difference in raw scores.

      Conclusions

      The MPAI-4 demonstrates satisfactory psychometric properties for use with individuals with CVA applying for interdisciplinary posthospital rehabilitation. DIF reveals clinically meaningful differences between CVA and TBI groups that should be considered in results at the item and subscale level.

      Key Words

      List of Abbreviations:

      ABI (acquired brain injury), CVA (cerebrovascular accident), DIF (differential item functioning), DTF (differential test functioning), MPAI-4 (Mayo-Portland Adaptability Inventory, 4th revision), PCA (principal component analysis), TBI (traumatic brain injury), TCC (test characteristic curve)
      IN THE MAJORITY of cases of stroke, infarction in the middle cerebral artery distribution results in circumscribed disabilities that are addressed through focused outpatient rehabilitation. However, in a substantial minority of cases, cerebrovascular accident (CVA) or multiple CVAs may affect a range of cognitive, motor, emotional, and behavioral functions requiring interdisciplinary inpatient and outpatient rehabilitation. Monitoring and evaluating progress in posthospital interdisciplinary rehabilitation after stroke also typically requires a more broadly based assessment approach. The Mayo-Portland Adaptability Inventory, 4th revision (MPAI-4), used extensively to measure progress and outcome in posthospital (ie, rehabilitation occurring after acute medical care and, in some cases, inpatient rehabilitation) traumatic brain injury (TBI) rehabilitation programs, may provide such an assessment.
      The MPAI-4 has a long history of psychometric development and validation.
      • Malec J.F.
      • Kragness M.
      • Evans R.W.
      • Finlay K.L.
      • Kent A.
      • Lezak M.
      Further psychometric evaluation and revision of the Mayo-Portland Adaptability Inventory in a national sample.
      • Malec J.F.
      • Lezak M.D.
      Manual for the Mayo-Portland Adaptability Inventory 2008.
      • Malec J.F.
      • Moessner A.M.
      • Kragness M.
      • Lezak M.D.
      Refining a measure of brain injury sequelae to predict postacute rehabilitation outcome: rating scale analysis of the Mayo-Portland Adaptability Inventory (MPAI).
      Most prior psychometric studies of the MPAI-4 have used mixed samples of individuals with acquired brain injury (ABI) in which the majority of subjects had TBI and a minority of subjects had other ABI diagnoses, most commonly, CVA. Kean et al
      • Kean J.
      • Malec J.F.
      • Altman I.M.
      • Swick S.
      Rasch measurement analysis of the Mayo-Portland Adaptability Inventory (MPAI-4) in a community-based rehabilitation sample.
      recently examined the psychometric properties of the MPAI-4 in a sample of 603 individuals, composed exclusively of patients with TBI on admission to home- and community-based rehabilitation programs. This study was concordant with previous analyses in mixed TBI and CVA samples and provided evidence of measurement reliability and construct validity both for the full MPAI-4 scale and its 3 subscales (ability, adjustment, participation). To our knowledge, no prior study has examined the psychometric functioning of the MPAI-4 in a sample comprised exclusively of patients with CVA.
      The MPAI-4 was developed using both classical and modern psychometric methods. The use of modern psychometric methods based on item-response theory, particularly Rasch analysis,
      • Bond T.M.
      • Fox C.G.
      Applying the Rasch model: fundamental measurement in the human sciences.
      has become standard practice in the development of rehabilitation metrics. Through such methods, a small set of items can be identified that represents the full range of a construct, such as outcome. Items are selected that have a predictable ordinal relationship to each other. In other words, each item (or, in the case of rating scales, item rating scale level) represents a specific level on the measure. Construction of measures in this fashion also allows for detailed assessment of item and test psychometric properties. For example, relevant to the current study, not only is the evaluation of reliability and construct validity needed, but the analysis of item and test bias in a sample of persons with CVA is also needed. Such analyses evaluate whether the MPAI-4 performs equitably across stroke subpopulations (eg, sex) and across rehabilitation populations (ie, TBI and CVA). Kean
      • Kean J.
      • Malec J.F.
      • Altman I.M.
      • Swick S.
      Rasch measurement analysis of the Mayo-Portland Adaptability Inventory (MPAI-4) in a community-based rehabilitation sample.
      reported no unexpected MPAI-4 item bias for sex, age, or chronicity in the analysis of a TBI sample. Similar analysis of item bias (ie, differential item functioning [DIF]) on the MPAI-4 has not been conducted in the CVA sample. Likewise, potential overall test bias to TBI or CVA populations (ie, differential test functioning [DTF]) has not been investigated.
      The objectives of the current study were: (1) to evaluate the measurement reliability and construct validity of the MPAI-4 in a sample consisting exclusively of patients with CVA using single parameter (Rasch) item-response methods, (2) to examine the DIF by sex within this CVA population, and (3) to examine DIF and DTF across TBI and CVA patient samples.

      Methods

      Participants

      The sample with CVA consisted of 861 adults admitted to a home- and community-based rehabilitation program with a CVA diagnosis; more specific information regarding the CVA was not available. Home- and community-based rehabilitation programs were located in 7 different geographically distinct cities in the continental U.S. The sample was 56% men with a mean age ± SD of 51.5±11.1 years (median, 52), who were admitted to the program an average ± SD of 114.4±329.7 days postevent (median, 47). The sample with TBI was the same as that used in Kean
      • Kean J.
      • Malec J.F.
      • Altman I.M.
      • Swick S.
      Rasch measurement analysis of the Mayo-Portland Adaptability Inventory (MPAI-4) in a community-based rehabilitation sample.
      and consisted of 603 individuals (72% men) with a mean age ± SD of 40.2±14.8 years (median, 39), who were admitted to the program an average ± SD of 461.6±1418.9 days postinjury (median, 81.0).

      Measure

      The MPAI-4
      • Malec J.F.
      • Lezak M.D.
      Manual for the Mayo-Portland Adaptability Inventory 2008.
      consists of 30 items selected to assess commonly occurring limitations after ABI. It is divided into 3 subscales: ability index, adjustment index, and participation index. The ability index is made up of 13 items to assess mobility, use of hands, vision, audition, dizziness, motor speech, verbal communication, nonverbal communication, attention/concentration, memory, fund of information, novel problem solving, and visuospatial abilities. The adjustment index includes 12 items for anxiety, depression, irritability/anger/aggression, pain and headache, fatigue, sensitivity to mild symptoms, inappropriate social interaction, impaired self-awareness, family/significant relationships, initiation, social contact, and leisure and recreational activities. Each item on both the ability and adjustment indices is rated on a 5-point scale: no problem (0); mild problem but does not interfere with activities, may use assistive device or medication (1); mild problem, interferes with activities 5% to 24% of the time (2); moderate problem, interferes with activities 25% to 75% of the time (3); severe problem, interferes with activities more than 75% of the time (4). More specific guidelines and anchors for the various levels of each item are provided in the MPAI-4 manual.
      • Malec J.F.
      • Lezak M.D.
      Manual for the Mayo-Portland Adaptability Inventory 2008.
      The participation index is made up of 8 items measuring initiation, social contact, leisure and recreational activities, self-care, independent living, employment, transportation, and money management. Most participation items are also rated on a 5-point scale similar to the scale for the ability and adjustment indices. However, ratings for the employment item are more specific, that is, representing variations of full-time, part-time, unpaid, and supported work and unemployment. Prior studies have demonstrated satisfactory internal consistency, construct validity,
      • Kean J.
      • Malec J.F.
      • Altman I.M.
      • Swick S.
      Rasch measurement analysis of the Mayo-Portland Adaptability Inventory (MPAI-4) in a community-based rehabilitation sample.
      • Bohac D.L.
      • Malec J.F.
      • Moessner A.M.
      Factor analysis of the Mayo-Portland Adaptability Inventory: structure and validity.
      as well as concurrent
      • Malec J.F.
      • Thompson J.M.
      Relationship of the Mayo-Portland Adaptability Inventory to functional outcome and cognitive performance measures.
      and predictive validity
      • Malec J.F.
      • Moessner A.M.
      • Kragness M.
      • Lezak M.D.
      Refining a measure of brain injury sequelae to predict postacute rehabilitation outcome: rating scale analysis of the Mayo-Portland Adaptability Inventory (MPAI).
      • Malec J.F.
      Impact of comprehensive day treatment on societal participation for persons with acquired brain injury.
      • Malec J.F.
      • Buffington A.L.
      • Moessner A.M.
      • Degiorgio L.
      A medical/vocational case coordination system for persons with brain injury: an evaluation of employment outcomes.
      for the full measure and its indices. The MPAI-4 has been found to be responsive to the effects of rehabilitation interventions.
      • Malec J.F.
      Impact of comprehensive day treatment on societal participation for persons with acquired brain injury.
      • Altman I.M.
      • Swick S.
      • Parrot D.
      • Malec J.F.
      Effectiveness of community-based rehabilitation after traumatic brain injury for 489 program completers compared with those precipitously discharged.
      • Constantinidou F.
      • Thomas R.D.
      • Scharp V.L.
      • Laske K.M.
      • Hammerly M.D.
      • Guitonde S.
      Effects of categorization training in patients with TBI during postacute rehabilitation: preliminary findings.

      Procedures

      On admission to the home- and community-based rehabilitation program, clinical evaluations were conducted by treating rehabilitation professionals and, after completion of the assessments, the MPAI-4 was completed by the treatment team based on consensus. The treatment team members rating the MPAI-4 were provided with written materials regarding the process of data collection and instructed to refer to the MPAI-4 manual for detailed rating instructions for each item.

      Analyses

      Analyses were performed on a deidentified dataset approved as exempt by the Indiana University Institutional Review Board. Rasch analysis creates a linear model of the latent construct based on a log-odds transformation of the probability of a specific response from a person in the sample. It allows a unified examination of several important aspects of measurement: reliability, construct validity, fit of items to the model, and differential functioning of items and tests. All Rasch analyses were conducted with WINSTEPS version 3.72.3a using Masters' partial credit model, and all other analyses were conducted with SPSS, version 18.0.b For a few items, combining levels is recommended on the scoring form. These item rating scale adjustments were made prior to analyses, resulting in a few items having a different number of levels than other items and requiring use of the partial credit model. The Rasch reliability statistics indicate the reproducibility of relative measure location on the modeled linear latent variable and are generated for modeled persons and items. Rasch separation is an expression of the reliability statistic on a scale from 0 to infinity and indicates the number of statistically different levels of performance that can be distinguished. To reflect this relationship, separation values are presented in parentheses after reliability statistics in the text. Conventional reliability and separation criteria of 0.8 and 2.0, respectively, were used in this study. Measures meeting these criteria allow for the detection of 3 distinct performance strata.
      Rasch infit and outfit statistics describe the fit of the items to the model and have a chi-square distribution and an expected value of 1. The deviation of items from expected model values indicates more or less variance in the data than predicted by the Rasch model. In this study, we examined infit and outfit statistics using rigorous criteria (1.0±0.3), which allowed for no more than 30% variation from the Rasch model to detect potential misfits in the polytomous data.
      • Smith A.B.
      • Rush R.
      • Fallowfield L.J.
      • Velikova G.
      • Sharpe M.
      Rasch fit statistics and sample size considerations for polytomous data.
      Final calibrations, DIF, and DFT were calculated using datasets that eliminated a small percentage (∼5%) of test protocols with evidence of abnormal test responding. Finally, principal component analysis (PCA) of the item residuals was conducted to determine if the subscale structure (ie, ability, adjustment, participation) established in previous studies of the MPAI-4 in ABI and TBI populations was applicable in CVA.
      Differences in functioning at the item level (DIF) within and across samples were calculated by contrasting log-odds estimates for item difficulty with person ability held constant. A t test was used to compare the statistical significance of contrast size divided by the joint SE of the 2 log-odds estimates. Items were considered to be differentially functioning if the log-odds estimates were both substantively (≥.50 logits) and statistically (P≤.05) significant. DTF is the examination of the impact of DIF across all items taken together. DTF was evaluated using 2 methods: least squares regression
      • Badia X.
      • Prieto L.
      • Linacre J.M.
      Differential Item and Test Functioning (DIF & DTF).
      and comparison of test characteristic curves (TCCs).
      • Wyse A.E.
      • Reckase M.D.
      A graphical approach to evaluating equating using test characteristic curves.
      Using the least squares method, the linear relationship between item parameter estimates for TBI and CVA samples was calculated. Using the TCC method, TCCs from the CVA and TBI samples were compared by computing the absolute value of the difference between raw score equivalents every .10 logits across the performance continua.

      Results

      Measure Calibration

      Initial calibration revealed person reliability/separation = .91/3.11. This person separation level indicates that 4 performance strata can be reliably identified. Item reliability/separation = 1.00/23.35. As a next step, 40 (∼5% of the sample) of the most statistically outlying cases were removed. Each of these cases demonstrated person infit or outfit >2.20, suggesting inattentive or otherwise invalid patterns of item responding. Removing these cases slightly improved person reliability/separation to .91/3.16 and item reliability/separation to 1.00/23.64.
      In order to evaluate the distinctiveness of rating levels for individual items, we examined the average ability estimate at each rating level for each item. Rating level assignment was disordered only for the residence (independent living) item. There did not appear to be a clear difference between level 1 (living without supervision but others have concerns about safety or managing responsibilities) and level 2 (requires a little assistance or supervision from others; 5%–24% of the time) in this sample, suggesting that these 2 levels are not clearly distinct to raters for this population. Although this lack of difference may be appropriate to consider in clinical use of the scale with stroke patients, no further adjustments to rating level categories were made prior to conducting fit analyses.
      The analysis of item fit at this step indicated 5 items with infit and outfit greater than 1.30 (infit/outfit in parentheses following the item): use of hands (1.71/1.71), motor speech (1.63/1.65), dizziness (1.45/1.60), vision (1.36/1.39), and verbal communication (1.37/1.37). Elimination of these 5 items from the scale had little effect on reliability and separation indicators. Person reliability/separation for the scale excluding these 5 marginal items were .91/3.11; item reliability/separation was 1.00/26.07. The point-measure correlations between these marginal items and the measure were at an acceptably moderate level, that is, use of hands = .41, motor speech = .46, dizziness = .33, vision = .49, and verbal communication = .54. The marginal items did not demonstrate notable patterns of rating category assignment or missing data. It should also be noted that setting the limits of infit and outfit from 0.7 to 1.3 is highly rigorous. Linacre,
      • Linacre J.M.
      What do infit and outfit, mean-square and standardized mean?.
      for example, suggests that items with infit and outfit between 0.5 and 1.5 typically contribute to measurement and that items with infit/outfit below 2.0 do not distort the measure.
      Targeting is the correspondence between the person measure and the item measure, that is, the degree to which items represent the full range of the construct being measured across persons. In this sample, targeting was excellent with the difference in means between person and item measures = |.11|, and the SD for the person measure was only slightly smaller (.71) than for the item measure (1.03).
      PCA of item residuals was used to determine if correlated residuals were present and suggestive of the subscale structure noted in previous analyses. Analysis indicated that the first component, the latent linear dimension created by the Rasch model, accounted for 37.6 Eigenvalue units (55.6%) of the variance in the data. The second, third, and fourth components constructed by the PCA met or exceeded the 2.0 Eigenvalue units considered to be chance variance. The Eigenvalue units and percentage of variance accounting for the second, third, and fourth were 3.8 (5.7%), 3.4 (5.1%), and 2.0 (3.0%) units, respectively. The second component contrasted cognitive items (memory, novel problem solving, fund of information) with physical items (mobility, use of hands, pain). The third component contrasted participation items (self-care, residence, mobility) with adjustment items (anxiety, sensitivity to mild symptoms, irritability, depression). Finally, the fourth component contrasted visual function items (visuospatial abilities, vision, dizziness) with verbal communication, motor speech functioning, and irritability. Using all items in the scale, reliability and separation indicators were lower for individual subscales than for the full measure as they were in analyses of subscale structure in other populations. For the ability index, person reliability/separation = .83/2.17 and item reliability/separation = 1.00/21.47. For the adjustment index, person reliability/separation = .84/2.28 and item reliability/separation = 1.00/26.54. For the participation index, person reliability/separation = .79/1.92 and item reliability/separation = 1.00/6.77.

      Differential Item Functioning

      To evaluate possible differences in item difficulty between men and women, log-odds estimates of item difficulty were calibrated separately for men and women and contrasted. No items demonstrated both substantive (≥.50 logits) and significant (P≤.05) differences in contrast size, which ranged from .00 to .26.
      To evaluate possible differences in item structure between TBI and CVA groups, 603 cases with a diagnosis of TBI from a prior study
      • Kean J.
      • Malec J.F.
      • Altman I.M.
      • Swick S.
      Rasch measurement analysis of the Mayo-Portland Adaptability Inventory (MPAI-4) in a community-based rehabilitation sample.
      were added to the sample with CVA. After an initial calibration of the combined sample, 73 cases with person infit or outfit >2.20 were dropped from the sample. Recalibration of the reduced sample resulted in overall measure indices that were, as expected, very similar to those obtained from the TBI and CVA groups separately, that is, person reliability/separation = .91/3.14 and item reliability/separation = 1.00/25.97. DIF analysis based on this second calibration revealed 7 items in which the measure value for the CVA group differed 0.5 logit or more from the value for the TBI group (CVA-TBI logit difference in parentheses): pain (.80), anger (.60), self-care (−.55), mobility (−.54), use of hands (−.63), audition (.51), and memory (.50).

      Differential Test Functioning

      Least squares

      A least squares regression model for the association between measure values for each item for the CVA group and for the TBI group was computed. As can be seen in figure 1, there is a high degree of association between the 2 measures (R=.92; R2=.85). The solid line in figure 1 is the least squares regression model; the dashed line illustrates the ideal perfect correlation (R/R2=1.00) between measures for the 2 groups. The 7 items that showed substantial differences at the item level are labeled in figure 1. Those above the lines are items that are associated with greater overall impairment for the CVA group and relatively less impairment for the TBI group; those below the lines are those associated with greater overall impairment for the TBI group and relatively less impairment for the CVA group.
      Figure thumbnail gr1
      Fig 1Measure values: CVA versus TBI. Abbreviations: Mem, memory; Mobil, mobility.

      Test characteristic curve

      To evaluate the effect of CVA and TBI item differences at the test level, we computed the absolute difference between the TCCs every .10 logits between −6.0 and 6.0. Figure 2 shows the TCCs for CVA and TBI. Figure 3 magnifies these differences and presents the absolute difference across the performance continua. The absolute difference ranged from .080 to a maximum of 3.70 observed (raw) score points. There was virtually no difference in the TCCs at the mean of items, because each TCC has its own zero-difficulty point set at the mean difficulty level. The largest differences (3–4 observed score points) were at approximately 2.0 and −2.0 logits. The SE of measurement is 5.4.

      Discussion

      This study describes basic psychometric properties of the MPAI-4 applied to patients with CVA. The construct validity and internal consistency of the MPAI-4 with this population appears very good. These psychometric features are equivalent to those obtained from the MPAI-4 in TBI and mixed samples. For both groups, the measure is able to separate 4 strata of disability (ie, minimal, mild, moderate, and severe). The performance of the MPAI-4 subscales (ability, adjustment, participation) is not as strong in stroke, TBI, or mixed samples but is adequate. Person separation values for the subscales indicate that 3 strata can be identified for each of the constructs measured (ie, mild, moderate, severe). It should not be surprising that the range of measurement for the subscales is not as great as for the full measure, because each subscale appears to primarily represent a level in the overall construct of outcome. Limitations apparent on ability items represent more severe overall disability; participation items represent hurdles for those with less overall disability; and adjustment items represent the midrange of overall disability (see Kean
      • Kean J.
      • Malec J.F.
      • Altman I.M.
      • Swick S.
      Rasch measurement analysis of the Mayo-Portland Adaptability Inventory (MPAI-4) in a community-based rehabilitation sample.
      ).
      Although the Rasch dimension accounted for more variance in the current CVA sample than in previous ABI and TBI samples, the analysis of item residuals demonstrated that statistically significant relationships between items remained outside of the variance accounted for by the Rasch dimension. Though the relationship between residuals is strictly a violation of the Rasch model, the tolerance of multidimensionality in a measurement instrument is a relative issue, driven by the measurement goal. Review of person reliability and separation statistics, which can be compromised by multidimensionality, suggests the measure demonstrates good construct validity in the CVA population. Multidimensionality can also compromise estimates of an individual's performance based on a few items, such as computer adaptive testing, though the relative brevity of the MPAI-4 makes such estimates unnecessary. Considering together the mitigation of these potential threats to validity and the good fit of items to the Rasch dimension suggests it is reasonable to recommend the MPAI-4 total score as a useful measure of global outcome after CVA. However, the evidence of multidimensionality and lack of faithfulness of item residual relationships to the existing subscale structure warrants further examination of subscale indices of the MPAI-4 in CVA populations. For example, future studies might include an investigation of item response by hemispheric lesion lateralization in CVA populations, as suggested by the contrast between verbal and nonverbal abilities in the fourth PCA component.
      The targeting or coverage of this stroke sample appeared very good. That is, the range of scores distinguishes well among individuals in the sample, and there is an absence of significant floor or ceiling effects. There was no substantive and significant DIF because of sex in the present sample. In previous analyses of patients with TBI, the audition item was biased against men, perhaps explained by more rapid longitudinal decline with age in the hearing levels in men.
      In the present sample, the greater mean and restricted variability of age of participants may explain the lack of difference. DIF because of age or chronicity was not examined because both these parameters were highly restricted in this sample.
      Various methods for examining DTF are emerging in the literature. We selected 2 approaches: 1 comparing item parameters using least squares regression and a second comparing TCCs. By both methods, performance of the MPAI-4 overall appeared very comparable between the TBI and CVA groups. The greatest discrepancy between raw scores for the stroke and TBI groups was 3.7 (see Fig 1, Fig 2), which is well within the estimated SE of measurement of 5.4. This means that the overall score on the MPAI-4 indicates a similar level of disability and limitation in participation whether applied to a person with stroke or a person with TBI.
      On the other hand, there were distinct differences between the level of overall disability represented by a handful of specific MPAI-4 items. In item-response theory, each item is selected to represent (with a degree of error) a specific level on the overall construct being measured. The level of measurement for the items in question differed between TBI and CVA samples. In figure 1, items below the lines represent relatively better overall functioning for CVA than for TBI patients; items above the lines represent better overall functioning for TBI than for CVA patients. Items with the most distinctive differences (per DIF analysis) are mobility, use of hands, pain, memory, self-care, anger, and audition. The implications of these findings are that in a CVA sample, motoric disability (mobility, use of hands) and associated limitation in self-care are often present, even among patients who are functioning relatively well overall; whereas, these limitations are more often associated with more severe overall disability in patients with TBI. In contrast, pain, memory problems, anger, and problems with audition tend to be present among higher functioning patients with TBI compared with those with CVA, and abnormal findings in these areas are associated with more severe overall disability for those with CVA.
      These apparent differences in item performance between these 2 populations are consistent with clinical experience. The presence of these differences supports the development of diagnostic specific reference norms for CVA and TBI populations for examination at the subscale and item level. Essentially equivalent DTF between the 2 groups suggests that the overall score for the MPAI-4 has a similar meaning whether applied to a patient with CVA or a patient with TBI. However, specific items do appear to represent different levels of outcome for these populations. This is likely to affect the subscale performance as well, because each subscale is composed of fewer items than the entire scale and more sensitive to the score on each item. As previously mentioned, PCA analysis also suggests a different subscale structure for the measure in the CVA population, possibly reflecting more specific disabilities related to more focal brain impairments. Because of its brevity, the MPAI-4 seems to be an unlikely candidate for conversion to a computerized adaptive testing format; however, computerized adaptive testing algorithms would need to be developed separately for CVA and TBI groups because of DIF between the 2 groups.

      Study Limitations

      Despite the large size of the sample, the restricted range in age and chronicity restricted the analyses.

      Conclusions

      This evaluation of the MPAI-4 in a large sample, exclusively composed of individuals with CVA, indicates that the MPAI-4 possesses satisfactory psychometric properties for the evaluation of progress and outcome in posthospital rehabilitation programs. The indication of overall level of disability and participation indicated by the total score for the measure is similar for patients with either CVA or TBI. Specific items, however, function differently within these 2 diagnostic groups and imply differing levels of overall disability consistent with the nature of brain impairment associated with each diagnosis. With knowledge of this DIF, the MPAI-4 should prove to be a useful measure for initial assessment and for monitoring progress and outcome of interdisciplinary posthospital rehabilitation for individuals with CVA.
      • a
        WINSTEPS; www.winsteps.com.
      • b
        SPSS; IBM Corp, 1 New Orchard Rd, Armonk, NY 10504-1722.

      References

        • Malec J.F.
        • Kragness M.
        • Evans R.W.
        • Finlay K.L.
        • Kent A.
        • Lezak M.
        Further psychometric evaluation and revision of the Mayo-Portland Adaptability Inventory in a national sample.
        J Head Trauma Rehabil. 2003; 18: 479-492
        • Malec J.F.
        • Lezak M.D.
        Manual for the Mayo-Portland Adaptability Inventory 2008.
        (Accessed July 10, 2012)
        • Malec J.F.
        • Moessner A.M.
        • Kragness M.
        • Lezak M.D.
        Refining a measure of brain injury sequelae to predict postacute rehabilitation outcome: rating scale analysis of the Mayo-Portland Adaptability Inventory (MPAI).
        J Head Trauma Rehabil. 2000; 15: 670-682
        • Kean J.
        • Malec J.F.
        • Altman I.M.
        • Swick S.
        Rasch measurement analysis of the Mayo-Portland Adaptability Inventory (MPAI-4) in a community-based rehabilitation sample.
        J Neurotrauma. 2011; 28: 745-753
        • Bond T.M.
        • Fox C.G.
        Applying the Rasch model: fundamental measurement in the human sciences.
        Lawrence Erlbaum, Mahwah2007
        • Bohac D.L.
        • Malec J.F.
        • Moessner A.M.
        Factor analysis of the Mayo-Portland Adaptability Inventory: structure and validity.
        Brain Inj. 1997; 11: 469-482
        • Malec J.F.
        • Thompson J.M.
        Relationship of the Mayo-Portland Adaptability Inventory to functional outcome and cognitive performance measures.
        J Head Trauma Rehabil. 1994; 9: 1-15
        • Malec J.F.
        Impact of comprehensive day treatment on societal participation for persons with acquired brain injury.
        Arch Phys Med Rehabil. 2001; 82: 885-894
        • Malec J.F.
        • Buffington A.L.
        • Moessner A.M.
        • Degiorgio L.
        A medical/vocational case coordination system for persons with brain injury: an evaluation of employment outcomes.
        Arch Phys Med Rehabil. 2000; 81: 1007-1015
        • Altman I.M.
        • Swick S.
        • Parrot D.
        • Malec J.F.
        Effectiveness of community-based rehabilitation after traumatic brain injury for 489 program completers compared with those precipitously discharged.
        Arch Phys Med Rehabil. 2010; 91: 1697-1704
        • Constantinidou F.
        • Thomas R.D.
        • Scharp V.L.
        • Laske K.M.
        • Hammerly M.D.
        • Guitonde S.
        Effects of categorization training in patients with TBI during postacute rehabilitation: preliminary findings.
        J Head Trauma Rehabil. 2005; 20: 143-157
        • Smith A.B.
        • Rush R.
        • Fallowfield L.J.
        • Velikova G.
        • Sharpe M.
        Rasch fit statistics and sample size considerations for polytomous data.
        BMC Med Res Methodol. 2008; 8: 33
        • Badia X.
        • Prieto L.
        • Linacre J.M.
        Differential Item and Test Functioning (DIF & DTF).
        Rasch Meas Trans. 2002; 16: 889
        • Wyse A.E.
        • Reckase M.D.
        A graphical approach to evaluating equating using test characteristic curves.
        Appl Psychol Meas. 2011; 35: 217-234
        • Linacre J.M.
        What do infit and outfit, mean-square and standardized mean?.
        Rasch Meas Trans. 2002; 16: 878