Analysis of probabilistic classification learning in patients with Parkinson's disease before and after pallidotomy surgery

Abstract

This study examined the characteristics of probabilistic classification learning, a form of implicit learning previously shown to be impaired in patients with basal ganglia dysfunction (e.g., Parkinson's disease). In this task, subjects learn to predict the weather using associations that are formed gradually across many trials, because of the probabilistic nature of the cue-outcome relationships. Patients with Parkinson's disease, both before and after pallidotomy, and age-matched control subjects, exhibited evidence of probabilistic classification learning across 100 training trials. However, pallidotomy appears to hinder the learning of associations most implicit in nature (i.e., weakly associated cues). Although subjects were most sensitive to single-cue associations when learning the task, there is evidence that cue combinations contribute significantly to probability learning. The utility of multiple dependent measures is discussed.

Figure 2

Cues used by subjects to predict the weather in a probabilistic manner (the probabilistic classification weather prediction task). One to three cards appeared on the screen during each trial. The subject was required to predict sun or rain based on these cues.

Figure 1

Coronal T2-weighted magnetic resonance imaging (MRI) scan showing a double electrode track for generation of a pallidotomy lesion away from the internal capsule and optic tract. Notice the mammillary body at the level of the pallidotomy lesion. Two lesions were performed to encompass the globus pallidus interna (GPI) (horizontal arrow). CSF, cerebral spinal fluid; MB, mammillary body; OT, optic tract.

Figure 3

(A) Probabilistic classification. Performance of normal, elderly control subjects on the 100-trial probabilistic classification (weather prediction) task is depicted as percent correct (number correct/total number of responses) across 20 trial blocks. Asterisks indicate performance significantly different from chance (dashed line, 50%; *P < .05, **P < .0005; one-sample t-test). These subjects exhibited a typical learning pattern, starting at chance and increasing by the second block to well above chance performance. Responses were considered correct if they corresponded to the more strongly associated outcome (sun or rain) on each trial. (B) Declarative memory. Subjects performed well above chance (25%) on an eight-item, four-choice questionnaire (shown as number correct/8), which assessed the appearance and characteristics of the task.

Figure 4

(A) Association strength–cue combinations. Control subjects exhibited above chance accuracy when using strongly associated cue combinations (75, 83, 86, and 89%), but not weakly associated combinations (63 and 67%). (B) Cue combination probability matching. The probability of a response in the presence of each cue combination (squares) and the actual programmed probability (Xs). Subjects significantly matched combination strengths but not to a very strong degree. (C) Individual cue probability matching. The probability of a response in the presence of each of the four individual cues compared with programmed probability, irrespective of whether cues appeared in a combination or alone. Normal subjects matched individual cue probabilities to a greater degree than combinations.

Figure 5

Performance of patients with Parkinson's disease before or after pallidotomy. (A) Probabilistic classification. Accuracy in 20 trial blocks is shown in patients that were tested before and/or after pallidotomy (All Patients, left panel), and a subset that were tested both before and after pallidotomy (Retest patients, right panel). Asterisks indicate performance significantly different from chance (50%; *P < .05). Postoperative patients began at chance and did not exceed chance performance until the last block of trials. In contrast, control subjects (Fig. 3A) exceeded chance by the second block of trials. Preoperative patients appear to spuriously exceed chance on the first block of trials, decline to chance levels, and then exceed chance by the last block of trials. This performance is not as robust as control subjects, and may be the result of motivational factors. (B) Declarative memory. Performance on an eight-item, four-choice declarative questionnaire regarding the weather prediction task is shown. All patients performed well above chance and comparable to control subjects (Fig. 3B) on this declarative memory test.

Figure 6

Performance on the first and last instance of each cue combination. To determine if the unusually high accuracy by preoperative patients in the first block of trials was spurious (Fig. 5A), we examined accuracy on the first and last instance of each cue combination. Control and postoperative patients began near chance, whereas preoperative patients were significantly above chance. Control subjects show a clear pattern of improvement, whereas both patient groups with Parkinson's disease show no evidence of improvement.

Figure 7

(A) Association strength—cue combinations. Postoperative patients with Parkinson's disease exhibited similar accuracy to preoperative patients when using strongly associated cue combinations (75, 83, 86, and 89%), but were somewhat impaired when using weakly associated combinations (63 and 67%). Compare with control subjects (Fig. 4A). (B) Cue combination probability matching. The probability of a response in the presence of each cue combination (squares) with the actual programmed probability (Xs). Subjects significantly matched combination strength but not to a very strong degree; moreover, probability matching was somewhat worse in postoperative patients. Compare with control subjects (Fig. 4B). (C) Individual cue probability matching. The probability of a response in the presence of each of the four individual cues compared with programmed probability, irrespective of whether cues appeared in a combination or alone.