Research Participants

Twelve patients with prefrontal cortex lesions (8 male and 4 female) and twelve age- and education-matched controls from the same community (8 male and 4 female) participated in the study (see Table 1 for participant characterization). The participants ranged in age from 54–81 years except for one 35 year-old patient and his age-matched control. There was no difference in age between patients and controls [F(1,22) = .08, ns; M = 65.08 and 63.67 years, respectively], or education [F(1,22) = .84, ns; M = 14.08 and 14.92 years, respectively]. Eleven participants in each group were right-handed; 1 patient and 1 control were left-handed.

Participant characterization

Lesion etiologies included stroke (n = 7), surgery for meningioma (n = 2), aneurysm (n = 1), cyst (n = 1), and arterio–venous malformation (n = 1). Lesions were localized in lateral prefrontal cortex in ten patients and in ventral frontal cortex in 2 patients (5 and 6). Seven patients had left-hemisphere lesions and the remaining 5 had right-hemisphere lesions (see Figure 1 for lesion reconstructions). Patients were excluded if they had lesions that extended beyond frontal cortex based on MRI and CT review. The average number of years post onset of injury was 10 years (range = 1–18 years). Exclusion criteria for all participants included prior history of drug or alcohol abuse, psychiatric disorders, or neurologic disorders. Patients were additionally screened for aphasia; all scored within normal limits on the Western Aphasia Battery (Kertesz, 1982).

Patients' lesions were reconstructed onto MRI templates. The far right image shows each patient's lesion projected onto the lateral surface of the brain. The numbers on the left side of the figure correspond to patients as listed in Table 1. Lesion reconstruction was unavailable for patient 7.

Testing was completed at the VA Northern California Health Care System in Martinez, CA. The study was approved by the Institutional Review Board, and all participants read and signed consent forms prior to participation.

Materials

The Word Context Test was administered as part of a larger executive functions battery (Delis-Kaplan Executive Functions System; Delis et al., 2001). It consisted of one practice trial and 10 target trials. On each trial, there was a series of five sentences that were presented in black print on a single page in a stimulus booklet. The examiner used a sheet of white paper to cover sentences so that only one sentence was revealed at a time. The first sentence provided very little specific information as to the identity of the correct response, and the subsequent sentences provided increasingly more specific contexts. The fifth and final sentence provided a fairly concrete context and made the identity of the correct response obvious to normal participants (see Table 2 for an example). The battery also included a verbal fluency task and the California Verbal Learning Test–II (CVLT–II; Delis et al., 2000). These tests were included in the current analysis because they provide conditions that assess retrieval from semantic knowledge and working memory, respectively. The CVLT–II is a list-learning task made up of 16 words, which are evenly distributed among four semantic categories (furniture, animals, ways of traveling, and vegetables). The verbal fluency task consisted of animal naming.

Sample control and patient performance on the Word Context Test: Target item “Eat”

Procedure

Participants were tested individually in a quiet testing room. Participants were first given instructions, which included the following: “Let's pretend you're in another country where they speak a different language. I have some words from this language, and I want you to figure out what the words mean.” Participants were told that they would see and hear a series of sentences with a made-up word and that they should guess what the word might mean after every sentence. The sentences were read out loud by the examiner and also appeared in the stimulus booklet. The examiner wrote down the participant's response after every sentence. Participants were never given feedback as to whether their responses were correct or not. They were simply told to continue giving the same answer if they thought they had the correct response. A practice trial was administered first, followed by 10 target trials. Each trial consisted of five sentences, which were revealed to the participant one at a time. Participants were able to see preceding sentences within a trial, so that they could refer back to them if necessary. A sample trial and responses are given in Table 2.

The CVLT–II and verbal fluency tasks were administered during the same testing session according to standardized instructions. The CVLT–II required participants to recall, in any order, a list of 16 words that were read by the examiner. This procedure was repeated across five trials. On the verbal fluency task, participants were asked to name as many items as possible in 90 s that belonged to the semantic condition.

Data Analysis

Both quantitative and qualitative analyses were done in an attempt to characterize performance on the Word Context Test. As a measure of overall performance, we calculated accuracy and the number of guesses needed to arrive at the correct response. Accuracy was calculated as the number of trials (out of 10) for which the participant provided the correct response by the final sentence of the trial. Total guesses were calculated as the number of guesses (out of 50) the participant made before arriving at and maintaining the correct response. Since the sentences within a trial become more and more concrete, this latter measure was used to determine whether frontal patients and controls differed in terms of how specific the information needed to be in order for them to arrive at the correct response. That is, a participant could get 10 out of 10 trials correct but have made a disproportionately large number of guesses, arriving at the correct response only at the final sentence of the trial. In order to better understand the source of inferencing deficits on this test, we also calculated a number of other scores. These included measuring whether participants' responses (1) made sense within the context of a given sentence; (2) made sense given the context of the preceding sentences of a trial; and (3) were consistent across sentences in a trial. These three scores overlap somewhat with the overall performance scores described above but were intended to further explore the source of inferencing deficits on this test.

The sentence context score was calculated to reflect how well a given response fit within the context of a particular sentence and thus whether a participant could integrate information within a single sentence. Responses were not scored on grammaticality but rather on how well they fit the sentence conceptually. The aim was to determine whether frontal patients could generate appropriate responses on a sentence-by-sentence basis, even if they were unable to integrate that information across sentences. Three trained raters assigned a score of 1–3 to each response, based on the following criteria: 1 = does not make sense, 2 = makes some sense, and 3 = makes very good sense. The total possible score per trial was 15. The raters were blind as to the participants' identity, and the interrater reliability on this measure was .89.

The integration score was calculated based on how well a given response fit the context provided by the preceding sentences. This score was computed in order to determine how well participants could integrate information across sentences, an ability we predicted would be impaired in frontal patients. Responses from only the 2nd, 3rd, and 4th sentences of each trial were rated, since the first sentence did not have any preceding context, and the final sentence provided very obvious information as to the correct response. The same blind raters scored these responses, and the interrater reliability for the integration score was .82.

The consistency score was calculated as the number of consecutively correct responses in the trial, divided by the number of the sentence (1–5) that first elicited the correct response. This ratio should be 100%, if a participant maintains the correct response on subsequent sentences within a trial once s/he arrives at it. The lower this ratio is, the more indicative of set-loss problems. Thus, this score provided an indication of how well participants were able to maintain set, an ability often found to be affected in frontal patients.

In order to successfully integrate information on the Word Context Test, participants may need to maintain previous cue information, and possibly their strategy, in working memory for mental manipulation. Performance on the integration score was therefore compared to trial one of the CVLT–II, to determine whether working memory played a role in inferential reasoning on this task. Data from the first trial of the CVLT–II were used in the current study as a measure of working memory, as this has been shown to be a reliable measure of verbal working memory and correlates with digit span (Delis et al., 1988).

Finally, as generation of verbal responses and word retrieval are processes integral to performance on the Word Context Test, the accuracy, total guesses, and sentence context measures were compared to semantic fluency, a measure of word retrieval ability (Baldo & Shimamura, 1998).