Knowledge of cognitive performance earlier in life is essential in order to characterize precisely the extent to which these abilities have declined when an individual is diagnosed as having a dementing illness. The National Adult Reading Test (NART) was developed by Nelson and O’Connell (1978) to estimate premorbid intellectual ability in patients suffering from intellectual deterioration due to dementia. The test consists of 50 words, graded in difficulty, whose pronunciation cannot be determined from their spelling. The ability to successfully read irregularly spelt words is relatively robust in the face of current cognitive impairment and is a sensitive marker of intellectual attainment. Because the NART relies on orthographic irregularities in the English language, the construction of analogues of the test in other languages is not simply a matter of translation of the test content. Rather, words in the target language that have comparable properties to those in the NART must be sought. A French adaptation of the NART – the fNART – was developed by Bovet (1991) and calibrated on a small French-speaking Swiss sample. In a sample of 30 nondemented subjects, number of words pronounced correctly correlated highly with WAIS-R (Wechsler, 1981) verbal and total IQ scores and less strongly with performance IQ (r = 0.43). Data available from an epidemiological survey undertaken in Geneva, Switzerland provided an opportunity to establish the measurement properties and construct validity of the fNART in a large sample unselected with respect to cognitive decline. In addition to the fNART, the survey incorporated a brief test battery assessing the domains of crystallized intelligence, memory and cognitive speed (see Christensen et al., 1999). An interview that enabled the diagnosis of dementia according to DSM IV criteria (the Canberra Interview for the Elderly, Social Psychiatry Research Unit, 1992), the Mini Mental State Examination (MMSE ; Folstein, Folstein and McHugh, 1975) and the Psychogeriatric Assessment Scales (PAS) (Jorm and Mackinnon, 1995 ; Mackinnon and Mulligan, 2001) were also administered. If the fNART measures intellectual ability, substantial correlations between it and the test battery would be expected. Further validation of the test was sought by exploring its relation with years of education. The stability of the fNART was assessed by comparing the scores of subjects with and without dementia, and by examining the relationship of fNART scores to an informant-based report of change in cognitive performance from earlier in life assessed in the PAS. If the fNART is stable in the face of cognitive deterioration, no between-group differences or association with reported cognitive change would be expected. Method – Subjects were randomly selected from residents of the canton of Geneva aged over 65 years. The analyses reported here were undertaken on a sample of 368 persons who gave codable responses to at least 90 % of the fNART items. They ranged in age from 65 to 94 years. Subjects were interviewed in their homes by trained lay interviewers. Results – Cronbach's alpha for the forty-item scale was high (0.89). The percentage of subjects correctly pronouncing words ranged from 7.3 % for « chamsin » to 96.7 % for « agenda ». Item response theory (IRT) models were fitted to the data. In a three-parameter model the value of the guessing (asymptote) parameter was vanishing small for all items. Accordingly, a two-parameter model was adopted. The discriminating power (slope) of items ranged considerably from 0.281 (rébus) to 1.192 (béotien). The average slope was 0.656. This corresponds to average factor loading of 0.528 (range 0.270 to 0.766.) The items measure a broad range of ability (mean threshold – 0.719, sd = 1.540). Most items, however, discriminate at moderate levels. The parameter values obtained in the current study were compared to those estimated in a French sample of persons at risk of dementia (see Mackinnon, Ritchie and Mulligan, 1999). The correlation between item pairs for slope and parameter estimates was 0.53 and 0.70 respectively. This indicated substantial concordance between the samples regarding the difficulty of the items, but some differences in the power of groups to differentiate ability. In particular, a small number of words that performed very well in the « at risk » sample showed more moderate discrimination in the current study. Scores on the fNART were correlated with measures of crystallised intelligence, memory and cognitive speed. All correlations were statistically significant. With all tests entered a regression equation the multiple correlation coefficient was 0.63. Mean fNART scores of those suffering from DSM IV dementia and those meeting only Criterion A (multiple cognitive deficits) were lower than those of subjects meeting neither set of criteria. However subjects in the first two groups were older than subjects in the undemented group and had significantly lower educational attainment. When these two factors were controlled in an analysis of covariance, the magnitude of the differences between the groups, while still overall significantly different, was substantially reduced. A similar pattern of results applied when psychometric measures of cognitive state – the MMSE and the PAS Cognitive Impairment Scale – were used instead of diagnostic categories. The partial correlations of the fNART with the MMSE and PAS cognitive impairment scale controlling for age and education were 0.25 (P < 0.01) and – 0.33 (P < 0.01) respectively. fNART scores did not differ between the sexes, nor were they significantly correlated with PAS Depression, Stroke or Behaviour Change scales. There was a small but significant correlation between the fNART and informant-assessed Cognitive Decline on the PAS. Discussion – This study demonstrated the excellent measurement properties of a French adaptation of the National Adult Reading Test in a large probability sample of elderly native speakers and provided the first large-sample evidence to support the validity of the fNART as a test of intellectual functioning relatively robust to dementia status. The negligible values of the pseudo-guessing parameters suggest that the goal of choosing words whose pronunciation is not susceptible to guessing has been achieved. The average item discriminability was high and the words used covered the spectrum of ability. The finding of substantial relationships of cognitive performance and educational attainment with fNART scores is important in validating the test as a measure of premorbid cognitive ability. The low correlations of the fNART with informant-based assessment of cognitive decline and age support the fNART as being relative robust to decline in ability. The relationships observed in this French adaptation are comparable to those reported for the English instrument (see Christensen et al., 1997). However, subjects meeting DSM IV criteria for dementia or Criterion A only had lower scores than other subjects. Decline in NART scores with dementia has been observed, particularly in moderate and severe cases. Given that the mechanism of the fNART is the same as the NART it is to be expected that while generally robust to current dementia status, some decline in performance will occur with the progression of the disease. The relationships between the fNART and PAS scales was remarkably similar to those reported by Jorm et al. (1995) in an English-speaking sample between the PAS and NART. Although small, the correlation between the fNART and the PAS Cognitive Decline scale might have been expected to be non-significant if the measure were truly stable in the face of intellectual deterioration. However this correlation is mirrored in the original English instruments and may reflect the higher risk of dementia in persons of lower intellectual ability. Conclusion – Further research is desirable to improve the precision of the calibration of the scale against the WAIS-R. Nevertheless, this study has demonstrated that the fNART is a reliable and valid method of assessing premorbid intellectual ability in French speakers.
