Association of Preoperative Visual Hallucinations With Cognitive Decline After Deep Brain Stimulation for Parkinson’s Disease

Deep brain stimulation (DBS) of the subthalamic nucleus (STN) (STN-DBS) or globus pallidus internus (GPi) can significantly improve the motor symptoms of Parkinson’s disease (PD) (1, 2). One concern with the neurosurgical procedure and chronic DBS, however, has been the risk of hastening the cognitive decline frequently associated with PD (3–6). Currently, there is no consensus on which presurgical characteristics elevate the risk of postoperative cognitive decline after DBS. Identifying these risk factors would inform a more personalized approach to care; in particular, about whether to proceed with surgery and, if so, how to help select the safest surgical approach, target, and stimulation strategy (7, 8).

Visual illusions or visual hallucinations (VHs) and impulse-control behaviors (ICBs) are two common nonmotor symptoms of PD, particularly after treatment with dopaminergic medications. In our clinic, we anecdotally observed that patients who reported a history of VHs experienced worse cognitive outcomes postoperatively, whereas those with a history of ICBs seemed to remain cognitively stable. Although VHs and ICBs are distinct phenomena, both may reflect unique neural network abnormalities or vulnerabilities and, as such, may provide prognostic information about cognitive outcomes after DBS.

In the case of VHs, prevalence in PD has been estimated at 22%-38% (9–12), and they occur on a spectrum from minor illusions to vivid hallucinations (13, 14). Although their etiology is likely complex (15–17), occurrence and progression in severity may be a behavioral marker of a more advanced and widespread neuronal pathology (18) and have been associated with cognitive changes (19–22) and reduced quality of life (13, 23, 24). This raises important questions about potential cognitive risks of DBS in patients with a history of VHs. Two studies incidentally described that psychosis-like symptoms in the immediate preoperative period appear to be associated with development of postoperative dementia (25, 26). To our knowledge, no study has systematically explored whether any history of VHs is associated with increased risk of post-DBS cognitive decline.

Similar to VHs, ICBs are fairly prevalent (in up to 20% of patients with PD), albeit associated primarily with dopaminergic therapy (27, 28). Unlike VHs, however, ICBs generally have not been linked to cognitive decline, although there have been reports suggesting differences in executive function between patients with and those without ICBs (29, 30). One study investigated relationships between ICBs and STN-DBS and found no cognitive differences at baseline or postoperatively between patients with and those without ICBs (31).

Motivated by our anecdotal clinical observations and to help clarify whether these common behavioral symptoms of PD and its treatment may reflect vulnerability to post-DBS cognitive decline, we used a retrospective clinical cohort to assess whether any preoperative history of VHs or ICBs was associated with clinically meaningful postoperative cognitive changes. As a secondary analysis, we determined whether cognitive outcomes in such patients differed depending on targeting of STN versus GPi.

Methods

Participants

We retrospectively reviewed electronic medical records of all patients with a diagnosis of PD who underwent DBS implantation in the STN or GPi between December 2012 and March 2018 (N=172), in accordance with a protocol approved by the University of Pittsburgh Institutional Review Board. Additional inclusion criteria included bilateral surgery, ongoing bilateral active stimulation after activation of the device, and the presence of both pre- and postoperative neuropsychological testing data. Exclusion criteria were clinical diagnosis of dementia at the time of preoperative evaluation, a Mini-Mental Status Examination score <24 at preoperative neuropsychological testing, and explant of the DBS device before postoperative neuropsychological testing. Lead implantation was performed as previously described utilizing either microelectrode recording or intraoperative MRI (32).

Chart Review

Demographic and relevant clinical data were collected for each subject by an abstractor (I.H.K.) who was blind to neuropsychological testing results. Medications at the time of both pre- and postoperative neuropsychological testing were reconciled, and levodopa equivalent daily dosages (33) were calculated for medications prescribed for motor symptoms. The preoperative number of nonmotor medications and Charlson Comorbidity Index score (34) were used as measures of medical comorbidity. Neurology and neurosurgery clinical notes were reviewed to confirm that each patient was receiving bilateral stimulation throughout the postoperative course, including during postoperative neuropsychological testing.

History of VHs was defined as documentation of any history of a visual misperception or hallucination outside the setting of acute delirium before DBS. History of ICBs was defined as documentation of a notable increase from baseline in impulsive or repetitive behaviors, typically related to shopping, sexual drive, or gambling. In addition to reviewing relevant medical and neuropsychological notes, we also utilized the Epic Systems, (Verona, WI) search function for the terms “hallucination,” “VH,” “illusion,” “shadow,” “impulse,” “compulsive,” “ICD,” “shopping,” and “gambling” to reduce the likelihood of missing the documentation of any such occurrences.

Neuropsychological Testing

All patients but one were assessed in our clinic. Before analyzing the data, we decided to utilize only a subset of tests from the full batteries with the goal of assessing all major cognitive domains but minimizing the number of statistical comparisons to decrease the risk of type I errors. On the basis of previous reports (35–37), we ultimately included 15 different measures: the Trail-Making Test, Part B; the Color-Word Interference Test (Stroop analog from the Delis-Kaplan Executive Function System); the Wechsler Abbreviated Scale of Intelligence–II Matrix Reasoning and Similarities subtests; the Repeatable Battery for the Assessment of Neuropsychological Status Semantic Fluency test, List Learning test, List Recall test, Figure Copy test, Figure Recall test, and Digit Span Forward test; the Boston Naming Test; the Wechsler Test of Adult Reading; the Controlled Oral-Word Association test; and the Wechsler Adult Intelligence Scale Digit Span Backward and Picture Completion subtests. These measures were ultimately consolidated into five cognitive domains (Table 1) to reduce the number of statistical comparisons and provide more clinically relevant results. Some patients were unable to complete all measures for various reasons (e.g., fatigue, sensoriperceptual problem), but all patients completed at least 13 of the measures both pre- and postoperatively, with the exception of two patients preoperatively and three patients postoperatively. One patient’s scores for the Digit Span Forward and Digit Span Backward subtests were excluded because of poor performance that was judged at the time of assessment by our neuropsychologist (L.C.H.) to reflect hearing impairment and not cognitive ability. Raw scores were converted to psychometric T scores (mean=50, standard deviation=10) to control for normative variables (i.e., age, sex, and level of education) and to facilitate direct comparison between pre- and postoperative scores.

Statistical Analysis

The distributions of continuous variables were checked for normality, and they were compared with paired or two-sample t tests, as appropriate, or Welch’s test for data with unequal variances. The chi-square test was used to compare categorical variables. For direct statistical comparison between descriptors of VH and ICB groups only, three patients who had a history of both symptoms were excluded. Mean preoperative and postoperative T scores were estimated for each subject individually and by cognitive domain according to the respective items in that domain (Table 1). To quantify changes in each domain, we then subtracted the mean preoperative T score from the mean postoperative T score to generate a ΔT score. Finally, associations between neuropsychiatric variable (VHs or ICBs) and cognitive domain were evaluated using linear regression. Covariates for the regression models initially included duration between neuropsychological testing and surgical target, and preoperative cognition-by-domain was added as a third covariate as a secondary analysis (i.e., preoperative mean T score for executive function was included as a covariate for the analysis of postoperative change in executive function, and so forth for each of the five cognitive domains). Significance level was set as α=0.05. The analytic plan to use linear regression models with covariates to control for surgical target and duration between neuropsychological testing was determined a priori, and analyses were performed with SAS, version 9.4 (SAS Institute, Cary, N.C.).

Results

Cohort Description

Chart review identified 54 patients out of a total operative sample of 172 (31%) who met entry criteria. From these 54 patients, we identified 15 (28%) with documented preoperative VHs (with illusions, N=2; hallucinations, N=13) and 12 (22%) with documented preoperative ICBs. These numbers include three (6%) patients with a documented history of both VHs and ICBs. The cohort overall was fairly homogeneous, consisting predominantly of right-handed, middle-aged, well-educated, Caucasian males who had been diagnosed with PD just under a decade before surgery (Table 2). Medical comorbidity was otherwise low. Mean time until follow-up neuropsychological testing was 1.4 years (median=0.77 years) with some variability.

Patients with a history of VHs or ICBs were similar at preoperative baseline to those without such a history (Table 2), with the following exceptions: Patients with a history of ICBs strongly trended toward being younger (62.8 years versus 67.6 years, p=0.053); were taking more PD medications for their motor symptoms (3.4 versus 2.6, p=0.045), with associated higher total levodopa equivalent daily dosages (1,447 mg versus 993 mg, p=0.019); had a greater improvement in preoperative Unified Parkinson’s Disease Rating Scale, part III (UPDRS-III), scores in the medicated (on) versus unmedicated (off) condition (54.4% versus 43.8%, p=0.042); and were less likely to have had STN as their DBS target (50% versus 86%, p=0.016).

Baseline Cognition

At preoperative neuropsychological testing assessment, on average, patients with a history of VH or ICB performed indistinguishably in each cognitive domain. (A table [Table S1] describing patient data is available as an online supplement to this article.) When analyzed by surgical target, baseline testing revealed worse performance by the GPi subcohort as compared with the STN subcohort on measures of executive function (40.87 versus 49.99, p=0.0001), memory (38.95 versus 46.20, p=0.011), and visuospatial cognition (40.06 versus 49.43, p=0.0002) (see Table S2 in the online supplement).

Analysis and Linear Regression Models

Changes in cognition in entire sample.

Analysis of the entire cohort’s change between pre- and postoperative neuropsychological testing without controlling for covariates demonstrated significantly worsened performance in executive function (ΔT=−3.64, p=0.0001) and language measures (ΔT=−2.05, p=0.001) (see Table S3 in the online supplement).

Effects of target and duration between follow-up on cognitive decline.

We used linear regression models to determine whether duration between pre- and postoperative neuropsychological testing or surgical target significantly contributed to ΔT scores for each cognitive domain. We found that duration between neuropsychological testing, but not DBS target, significantly contributed to changes in cognitive performances (Table 1). Specifically, increased duration between neuropsychological testing was associated with mild worsening in executive function (b=−2.22, p=0.004) and memory (b=−3.12, p=0.002).

Effects of history of VHs or ICBs.

Next, we sought to determine whether a history of VHs or ICBs was associated with specific cognitive changes at follow-up neuropsychological testing. We again used linear regression models and added a term for neuropsychiatric history as an additional covariate, along with duration between neuropsychological testing and surgical target. As shown in Table 3, history of VHs explained approximately one-half of a standard deviation decrement in performance in attention (b=−4.04, p=0.041) and executive function (b=−4.24, p=0.021). Conversely, history of ICBs was not associated with any statistically significant post-DBS cognitive changes.

To determine whether the identified associations were driven by those patients with frank VHs, we excluded the two patients with visual illusions and repeated the analysis (Table S4 in the online supplement). The significant association between history of VHs and worsened executive function was preserved (b=−4.79, p=0.013), whereas the association with worsened attention became nonsignificant.

Next, we included preoperative cognition as an additional covariate in our models to determine its influence on outcomes. The results were largely unchanged (see Table S5 in the online supplement; for comparisons including or excluding patients with visual illusions, see Table 3 and Tables S4 and Table S5 in the online supplement), although the association between impaired postoperative attention and history of VHs became only borderline significant (p=0.050).

Finally, to further explore the impairments in attention and executive function associated with a history of VHs, separate linear regression models were carried out for each of the eight cognitive tests that were used to define the domains (Table 1). In contrast to the domain-based analysis, a history of VHs was not significantly associated with worsened performance on any individual measure (Table S6 in the online supplement).

Discussion

Using data from a retrospective surgical cohort at an academic medical center, we demonstrated that any preoperative history of VH was significantly associated with worsening of postoperative cognitive performance in attention and executive function in patients with PD receiving DBS. We did not find any such association with a preoperative history of ICBs. Surgical target (STN or GPi) did not affect cognitive outcomes, but increased duration between pre- and postoperative neuropsychological testing was consistently associated with mildly worsened performances on attention and memory tasks.

The notion that neuropsychiatric history might influence DBS outcomes has support in the literature. For example, one study found that patients with active preoperative diagnoses of mood or anxiety disorders had greater improvements in mood after DBS (38). A unique aspect of our approach, however, was to include any history of VH or ICB, regardless of whether the symptoms were present at the time of DBS surgery. Previous studies of these nonmotor symptoms generally categorized the presence of symptoms only if they occurred during the immediate presurgical period (e.g., 1 week previously for the UPDRS, section I, and 1 month previously for The Neuropsychiatric Inventory Questionnaire). For example, Funkiewiez et al. (39) utilized the thought disorder question in UPDRS section I (question 2) and identified only three of 70 patients (4%) preoperatively who had experienced hallucinations (or delusions) in the week before surgery. We reasoned that susceptibility to these symptoms, even if caused by dopaminergic medications and resolved after cessation of the causative agents, may imply underlying vulnerabilities in specific brain networks that could portend vulnerabilities for cognitive changes after anesthesia, neurosurgery, chronic brain stimulation, and associated medication changes.

This approach allowed us to identify 15 out of 54 patients (28%) who had ever experienced preoperative VHs and may explain why we were able to detect robust changes in cognition associated with a history of VHs that had been only hinted at in previous reports. Aybek and colleagues (25), for instance, found that all five (9%) of their initial 57 patients who had endorsed psychosis-like symptoms on their preoperative UPDRS, part I (item 2, score ≥2), were diagnosed with dementia 3 years later. Similarly, Perriol et al. (26) noted in their study of 58 patients that three out of the four who experienced postoperative cognitive complications had a “history of dopaminergic psychosis,” but this was not further characterized or discussed. Thus, although the use of retrospective chart review and lack of validated instruments to assess for history of VHs (and ICBs) may be considered a limitation, our approach may have identified patients with such a phenotype with greater sensitivity, thereby providing greater statistical power to identify cognitive declines associated with a history of VHs that only had been suggested previously.

Although this finding is concerning and, if replicated, might inform preoperative counseling, risk assessment, improved patient selection, and perhaps even remediation strategies, it is also important to stress that, on average, performance worsened only by less than one-half of one standard deviation in two of the five cognitive domains assessed. Depending on an individual patient’s burden of motor symptoms and overall clinical context, such outcomes may be considered an acceptable tradeoff, and a history of VHs alone does not appear to represent an absolute contraindication to DBS.

One potential concern with our approach was the grouping of patients with a history of illusions together with patients with a history of frank VHs, as considerably less is known about the prognostic implications of milder visual phenomenon (13). The small number of patients with a history of only illusions precluded analysis of these patients independently. Instead, we performed a sensitivity analysis by removing these patients and rerunning our linear regression models only with those patients with a documented history of frank VHs. We were surprised that removing patients with illusions eliminated the association between VHs and postoperative decline in attention (Tables S4 and S5 in the online supplement). This sensitivity analysis suggests that the variance is not explained solely by those patients with frank VHs and is consistent with other studies, indicating that even more minor visual phenomena may have clinical and prognostic significance (13, 23).

Another potential limitation was the possibility that differences in baseline cognition between the groups may explain the detected associations. At preoperative testing, however, the cognitive performance across groups was statistically indistinguishable (Table S1 in the online supplement). To further test the rigor of our approach, we repeated the analyses and included preoperative cognition-by-domain as an additional covariate. This sensitivity analysis did not change the outcome, reinforcing that differences in baseline cognitive performance did not explain our observations.

In the case of history of ICBs, we did not detect a significant post-DBS change in cognition, consistent with a previous report (31). This is compatible with the idea that ICBs may be attributable in part to iatrogenic overstimulation of the mesocortical and mesolimbic pathways as a consequence of pharmacological attempts to sufficiently augment the nigrostriatal (motor) system, which is then relieved by post-DBS dopaminergic medication reductions. Indeed, in our cohort, a history of ICBs was associated with both a greater preoperative burden of dopaminergic medication and a greater postoperative reduction (Table 2).

We also wondered whether surgical target might differentially influence results based on neuropsychiatric history, particularly given ongoing debate about putative cognitive differences associated with STN or GPi targeting (5, 40). We found no support for this idea. However, a major caveat is the nonrandom fashion in which surgical target was determined for each patient. To better understand what bias this might introduce, we subsequently reviewed the documented rationale for GPi targeting in each case. In eight of 12 cases (three of six ICB patients, five of six non-ICB patients), GPi was chosen because of identified preoperative concern about baseline cognition. Thus, if poorer baseline performance implied increased vulnerability and greater risk for further deterioration, then we might have found GPi stimulation to be associated with worsened results in our linear regression model. The fact that cognitive worsening was not explained by the choice of GPi as target could be interpreted as indirect evidence that GPi may be relatively cognitive sparing. Conversely, poorer preoperative performance in some cognitive domains in the GPi subcohort might have resulted in a floor effect that obscured our ability to detect more significant changes between targets. The smaller size of our GPi subcohort, however, combined with our methodological design, ultimately limits any conclusions.

We acknowledge several additional limitations. Although VHs are the most common form of psychotic symptoms in PD, we did not attempt to identify whether patients had delusions or other forms or hallucinations. Furthermore, there may have been unknown bias related to capturing only the patients who returned for follow-up neuropsychological testing, and the homogeneity of the demographics of the cohort and the academic setting may limit generalizability of our results to other populations. Another potential limitation is our lack of outcome data regarding motor symptoms and functional outcomes. As a retrospective study of a clinical population, we found that most patients did not have a documented postoperative UPDRS assessment to quantify motor or functional outcomes near the time of follow-up neuropsychological testing, if at all. As such, we could not adequately include this variable in our analysis, although we do note that postoperative decreases in dopaminergic medications—usually attributable to DBS-related improvement in motor symptoms—were not statistically different among our groups (Table 2), and past reports have not found a correlation between motor and cognitive outcomes (26). Nevertheless, we cannot exclude the possibility that motor or functional outcomes may have been different among the groups. Finally, and as noted earlier, identification of a history of VH or ICB was not based on any validated assessments. It is possible that our chart review failed to identify all patients who had ever experienced these symptoms or inaccurately identified patients who did not. Either possibility would result in a diluting of statistical power, and so the fact that we were able to identify significant effects further reinforces the robustness of our findings.

In conclusion, any history of VH during the course of PD may convey prognostic information about post-DBS cognitive risk. Further research is needed to determine whether these findings can be replicated in a larger, more diverse cohort for which motor and functional outcomes (i.e., scores on the UPDRS parts I–IV) are also available for inclusion in the analysis. With few, if any, validated predictors of DBS outcomes beyond preoperative response to levodopa, routine expansion of pre-DBS assessment to a more comprehensive survey of potential neuropsychiatric symptoms may be an important step toward optimizing DBS-related decision making for each individual patient.