Postictal Psychosis in Partial Epilepsy:A Case–Control Study

Kenneth Alper, MD,1,2 Ruben Kuzniecky, MD,1 Chad Carlson, MD,1 William B. Barr, PhD,1

Charles K. Vorkas, BA,1 Jignasa G. Patel, MD,1 Angela L. Carrelli, MD,1 Karen Starner, RN,1

Peter L. Flom, PhD,3 and Orrin Devinsky, MD1,2,4

Objective: Divergent findings among prior studies on correlates of risk for postictal psychosis (PIP) suggest the value of acontrolled study involving a relatively large number of patients.Methods: The study population consisted of a consecutive series of 59 patients with partial epilepsy and a history of PIP, and94 control patients with partial epilepsy and no history of PIP evaluated as inpatients with video-electroencephalography. Thegroups did not differ significantly regarding demographic features. Exact tests yielded a subset of variables and a tentativeinterpretation that were evaluated further utilizing principal components analysis and logistic regression.Results: PIP was associated with extratemporal versus temporal (p � 0.036) or undetermined (p � 0.001) localization of seizureonset, bilateral interictal epileptiform activity (p � 0.017), secondary generalization (p � 0.049), and history of encephalitis(p � 0.018). Interictal slow activity was more frequently absent in control patients (p � 0.045). PIP was associated with familyhistories of psychiatric disorders (p � 0.007) and epilepsy (p � 0.042), which themselves were significantly intercorrelated (r �0.225; p � 0.006). Age of onset or duration of epilepsy and lateralized electroencephalographic or magnetic resonance imagingasymmetries did not differ significantly between control and PIP groups. The analysis indicated four underlying domains of riskfor PIP: ambiguous/extratemporal localization, family neuropsychiatric history, abnormal interictal electroencephalographic ac-tivity, and encephalitis. Each unit increase on a simple additive scale composed of 9 dichotomous independent variables mul-tiplied the odds ratio for PIP by 1.71 (95% confidence interval, 1.36–2.15; p � 0.0001).Interpretation: PIP in partial epilepsy is associated with relatively broadly and bilaterally distributed epileptogenic networks,genetic determinants of psychiatric disorders and seizures, and encephalitis.

Ann Neurol 2008;63:602–610

Postictal psychosis (PIP) is a common and clinicallysignificant problem, and an important neuropsychiatricparadigm. It is a safety concern in patients undergoingnoninvasive and invasive video-electroencephalographic(VEEG) monitoring. PIP is defined by the occurrenceof psychotic symptoms after a seizure or seizure clusterwithout alternative explanatory causative factor.1 Psy-chosis can occur immediately after a seizure, or with atime offset between seizures, and the subsequent occur-rence of psychotic features that has been termed a lucidinterval, with duration of up to a week.

Two studies have reported on the incidence of PIPin VEEG monitoring studies and found rates of 6.4%2

and 6.7%,3 respectively. Although interictal psychoseswere considered the predominant psychotic presenta-tion in epilepsy, PIP was more common than interictalpsychosis in some recent series.4 PIP is associated withprogression to chronic psychotic disorders.5 Like other

psychotic syndromes, PIP is associated with increasedmortality,1,3 which is mediated by epilepsy severity, aswell as an increased incidence of suicide, particularlyinvolving violent means.6

Several neurological factors are reported to increaserisk for development of PIP, including temporal lobeepilepsy, bitemporal seizure foci, causative processes as-sociated with bilateral limbic lesions (eg, encephalitis,head trauma), and a tendency to cluster preceding themanifestation of psychotic symptoms.1,2,7–9 PIP hasalso been associated with a family history of mood dis-orders.3

This study reports on a consecutive series of 59 pa-tients with partial epilepsy and a history of PIP, and 94control patients with partial epilepsy and no history ofPIP evaluated as inpatients with concurrent VEEG aspossible candidates for epilepsy surgery. The PIP andcontrol groups were compared on variables in the do-

From the 1Department of Neurology, Comprehensive EpilepsyCenter; 2Department of Psychiatry, New York University School ofMedicine, New York, NY; 3Independent statistical consultant; and4Department of Neurosurgery, New York University School ofMedicine, New York, NY.

Received Jun 26, 2007, and in revised form Nov 16. Accepted forpublication Dec 21, 2007.

Published online in Wiley InterScience (www.interscience.wiley.com).DOI: 10.1002/ana.21341

Address correspondence to Dr Alper, NYU Comprehensive EpilepsyCenter, New York University School of Medicine, 403 East 34thStreet, 4th Floor EPC, New York, NY 10016. E-mail: kra1@nyu.edu

602 © 2008 American Neurological AssociationPublished by Wiley-Liss, Inc., through Wiley Subscription Services

mains of neurological and psychiatric family history,magnetic resonance imaging (MRI) abnormality, clini-cal features of seizures and lateralization and localiza-tion of VEEG seizure onsets, and interictal epilepti-form activity (EA) and slowing.

Patients and MethodsSample Characteristics and Inclusion/ExclusionCriteriaThe Institutional Review Board of the New York UniversitySchool of Medicine approved this study. Subjects were 153patients with partial epilepsy undergoing continuous VEEGmonitoring for evaluation as possible candidates for epilepsysurgery at the New York University Comprehensive EpilepsyCenter (Table 1). The 59 index patients had a history of PIPand were compared with 94 control patients without a his-tory of PIP. The first author (K.A.), a psychiatrist, routinelyevaluates patients undergoing VEEG monitoring, and evalu-ated each patient in this study in an interview that averaged1 hour in duration. The patient groups were selected retro-spectively based on review of the initial psychiatric consulta-tion report, which was systematically retained for every pa-tient seen. The study sample consisted of all patients meetingstudy inclusion and exclusion criteria (see later) selected con-secutively from a total of 1,340 psychiatric consultations per-formed (by K.A.) in the epilepsy unit from 1999 to 2003.

The sample was selected blind to the neurological history,family history of seizures, and VEEG and MRI results.

Indications for psychiatric evaluation at the New YorkUniversity Comprehensive Epilepsy Center include patientswho are surgical candidates without any other presentingpsychiatric or behavioral issues. All such patients are seenroutinely. Those patients who further met the inclusion andexclusion criteria without any history of PIP at the time ofthe evaluation or subsequent follow-up constituted the con-trol group. The PIP patients presented with the onset of PIPduring or immediately preceding (within 24 hours) the ini-tiation of the VEEG monitoring study. The attribution ofpsychosis was made by direct observation of clinical featuresof psychosis as defined by Diagnostic and Statistical Manualof Mental Disorders, Fourth Edition,10 namely, hallucina-tions, delusions, or grossly disorganized behavior. Historieswere obtained with the verification of at least one reliableobserver/informant such as a parent or partner.

Inclusion criteria that were applied to all patients in thestudy were a diagnosis of partial epilepsy and fluency in En-glish. Exclusion criteria for all patients were: being adopted,any comorbid epilepsy diagnosis other than partial epilepsy, adiagnosis of pervasive developmental disorder or autism,mental retardation, schizophrenia, schizoaffective, or mooddisorder with psychotic features. Patients with brain tumorsor receiving steroid medication were also excluded. Prior orcurrent psychiatric treatment for nonpsychotic disorders

Table 1. Demographic, Neurological, and Family History Variables

Variables Control Group PIP Group p

Demographics

N 94 59

Age, yra 35.0 � 12.4 36.2 � 10.4 0.494

Female sex, n (%) 53 (56.4) 29 (49.2) 0.409

Education, yra 13.3 � 4.6 13.2 � 6.7 0.886

Neurological history

Right-handed, n (%) 72 (76.6) 45 (76.2) 0.999

Age at onset of epilepsy, yra 16.0 � 11.8 17.0 � 11.2 0.614

Years with epilepsya 19.3 � 12.7 19.5 � 12.1 0.903

History of febrile seizures, n (%) 12 (12.9) 9 (15.5) 0.639

History of encephalitis, n (%) 4 (4.3) 10 (17.0) 0.018b

History of traumatic brain injury, n (%) 23 (24.5) 17 (28.8) 0.575

Family history

Seizures, n (%) 14 (14.9) 17 (28.8) 0.042b

Any psychiatric disorder, n (%) 30 (31.9) 32 (54.2) 0.007b

Mood, n (%) 18 (19.2) 20 (33.9) 0.054

Anxiety, n (%) 3 (3.2) 6 (10.2) 0.089

Psychotic, n (%) 1 (1.1) 4 (6.8) 0.073

Alcohol use, n (%) 13 (13.8) 18 (30.5) 0.022b

aUnpaired t test. All other comparisons are computed using the Fisher exact test.bp � 0.05.PIP � postictal psychosis.

Alper et al: PIP in Partial Epilepsy 603

with medications other than antipsychotics were notgrounds for exclusion for either the control or PIP patientgroups.

Inclusion criteria for the control patients were that theonly reason stated on the consult request was that the patientwas a surgical candidate but had no acute behavioral indica-tion for the consultation. Exclusion criteria for control pa-tients were any prior or current treatment with antipsychoticmedication for any reason and any history suspicious for aprior psychotic episode, such as psychiatric hospitalizationsor episodes that appeared to possibly involve hallucinations,delusions, or grossly disorganized behavior.

For inclusion criteria regarding PIP patients, psychosis wasdefined in the Diagnostic and Statistical Manual of MentalDisorders, Fourth Edition10 as presence of hallucinations, de-lusions, or disorganized behavior. The criteria that Logsdailand Toone1 provided were used to classify a psychotic epi-sode as postictal: (1) emergence of psychotic state within 7days of a seizure or seizure cluster; (2) duration of psychoticepisode for at least 15 hours; and (3) no evidence of con-founding factors that could explain the psychotic state froma clinical cause other than PIP.

Neurological and Family Psychiatric History DataNeurological histories and family histories of seizures wereobtained from the inpatient admission form. The first authorobtained all family psychiatric histories from the patient or aknowledgeable family informant, or both, using the approachdescribed in a previous study from this center.3 The objectivewas to determine a lifetime prevalence in the patient’s familyof significant psychiatric disorders in any of four major di-agnostic categories, namely, mood, anxiety, psychotic, andalcohol use disorders in first- (ie, parents and siblings) andsecond-degree (ie, grandparents, aunts, and uncles) relatives.The psychiatric family history specifically included the rela-tive’s inpatient or outpatient treatment, use of psychophar-macological agents, substance abuse, mood and psychoticsymptoms, panic episodes, obsessions, compulsive behaviors,phobias, and suicide attempts.

The approach was adapted from the Family History Re-search Diagnostic Criteria (FH-RDC),11 which applies theFeighner/RDC diagnostic criteria12 to obtain a retrospectivediagnosis of relatives from the accounts of probands andfamily members. A relative’s case was regarded as positivewhen the disorder was significant and could be classified inat least one of four major psychiatric categories. The crite-rion for case significance was a history of either definite func-tional impairment or having been treated for the disorder.The specificity associated with an emphasis on more severecases was regarded as worth the trade-off of limited sensitiv-ity to detect mild cases.13,14 The diagnostic inclusion crite-rion was a lifetime prevalence of at least one FH-RDCmood, psychotic, or alcohol use disorder, or an RDC anxietydisorder.

Magnetic Resonance Imaging, Clinical Seizure, andVideo-Encephalographic DataClinical seizure and EEG data were extracted from epilep-tologists’ reports on VEEG studies that utilized standardscalp electrodes. The extracted variables included secondary

generalization or clustering, which was defined as two ormore seizures in a 24-hour period15; laterality and temporalversus extratemporal localization of seizure onsets were basedon EEG and clinical features, and the laterality of interictalEA and slowing. MRIs were classified as showing a focal orunilateral abnormality, bilateral abnormalities, or normal.Abnormalities included hippocampal sclerosis and develop-mental or vascular malformations.

Statistical AnalysisStatistical analysis was performed utilizing StatsDirect statis-tical software version 2.5.6 (StatsDirect, Cheshire, UnitedKingdom) and SAS version 9.1 (SAS Institute, Cary, NC).

UNIVARIATE/BIVARIATE ANALYSIS. The Fisher–Freeman–Halton (generalized Fisher exact) test16 was used to evaluatecomparisons between PIP and control patients on categoricalvariables as 2 � n contingency tables. For 2 � 2 tables, theFisher–Freeman–Halton test yields the same result as theFisher’s exact test. Differences between continuous variableswere evaluated using the unpaired t test with the exceptionof the variable for the total number of seizures during thestudy (Table 2), for which the Mann–Whitney U test wasused because of its strongly nonnormal distribution necessi-tating a nonparametric method.

MULTIVARIATE ANALYSIS. A subset of variables was se-lected from the initial univariate/bivariate analyses in Tables1 and 2 that appeared to relate to a tentative interpretationand account for the distinction between the control and PIPgroups. Principal components analysis (PCA) with varimaxrotation17 was used to further reduce the set of intercorre-lated univariate measures. Although PCA focuses on reduc-ing intercorrelated variables and is a mathematical techniquedistinct from factor analysis, which focuses on identificationof latent factors, this article follows the common practice ofdesignating principal components as “factors.” For logisticregression analysis, the use of factors enabled the conversionfrom dichotomous to continuous independent variables, al-lowing narrower confidence intervals around the odds ratios(ORs). Bivariate regression on a simple additive PIP riskscale was compared with multiple regression on factor scores.

ResultsUnivariate/Bivariate Analysis

DEMOGRAPHIC VARIABLES. As indicated in Table 1,the control (n � 94) and PIP (n � 59) groups do notdiffer significantly with respect to age (p � 0.494), sex(p � 0.409), or years of education (p � 0.886).

NEUROLOGICAL HISTORY. The two groups did not dif-fer with respect to age of onset (p � 0.614) or chro-nicity of epilepsy (p � 0.903). The PIP group had agreater frequency of a history of encephalitis (p �0.018), but the two groups did not differ with respectto traumatic brain injury (p � 0.575) or histories offebrile seizures (p � 0.639).

604 Annals of Neurology Vol 63 No 5 May 2008

FAMILY HISTORIES OF EPILEPSY AND PSYCHIATRIC DISOR-

DERS. A family history of seizures was more prevalentin the PIP group (p � 0.042). A family history of atleast one among any the four categories of psychiatricdisorders was more prevalent in the PIP group (p �

0.007). However, the tendency for greater prevalencein the PIP group was less strong or insignificant amongthe four individual categories of psychiatric disorders:mood (p � 0.054), anxiety (p � 0.089), psychotic(p � 0.073), and alcohol use (p � 0.022) disorders.

Table 2. Magnetic Resonance Imaging and Video-Encephalographic Variables

Variables Control Group PIPGroup

p

MRI abnormality

Left 31 (38.3%) 16 (30.2%)

Right 14 (17.3%) 9 (17.0%)

Bilateral 16 (19.8%) 11 (20.8%)

No abnormality 20 (24.7%) 17 (32.1%) 0.738

Clinical seizures

Total seizures during study 6.82 6.39 0.182b

Clustering 26 (33.3%) 25 (47.2%) 0.144

Secondary generalization 39 (50.0%) 36 (67.9%) 0.049d

Seizure onset

Laterality (L vs R)

Left 37 (47.4%) 22 (41.5%)

Right 19 (24.4%) 15 (28.3%)

Bilateral independent 10 (12.8%) 8 (15.1%)

Undetermined 12 (15.4%) 8 (15.1%) 0.911

Localization (temporal vs extratemporal)

Temporal 53 (68.0%) 25 (47.2%)

Extratemporal 15 (19.2%) 21 (39.6%)

Undetermined 10 (12.8%) 7 (13.2%) 0.036d

Undetermined onset

One or more seizures with both laterality and localizationundetermined 24 (30.8%) 32 (60.4%) 0.001d

Interictal epileptiform EEG activity

Left 41 (43.6%) 15 (25.4%)

Right 14 (14.9%) 4 (6.8%)

Bilateral 29 (30.9%) 31 (52.5%)

None 10 (10.6%) 9 (15.3%) 0.017d

Interictal EEG slow activity

Left 23 (24.5%) 19 (32.2%)

Right 10 (10.6%) 6 (10.2%)

Bilateral/generalized 13 (13.8%) 14 (23.7%)

None 48 (51.1%) 20 (33.9%) 0.152e

aMagnetic resonance imaging (MRI) reports were available for 81 of 94 control patients and 53 of 59 patients with postictal psychosis(PIP). bMann–Whitney U test. All other comparisons in this table were computed using the Fisher–Freeman–Halton exact (generalizedFisher exact) test. cSixteen control and 6 PIP patients did not produce seizures during the video-electroencephalographic (VEEG) study;therefore, comparisons regarding clinical seizures and seizure onsets were made on 78 control and 53 PIP patients, respectively. dp �0.05. eThe 2 � 2 contingency table combining all forms of slowing was significant (p � 0.045) and indicated that “none” was morecommon in control patients.

Alper et al: PIP in Partial Epilepsy 605

In addition, correlations between family histories ofseizures and of psychiatric disorders were examined. Afamily history of seizures correlated significantly with afamily history of any psychiatric disorder (r � 0.225;p � 0.006) and mood (r � 0.206; p � 0.011) oranxiety (r � 0.218; p � 0.007) disorders, but not psy-chotic (r � �0.002; p � 0.976) or alcohol use (r �0.107; p � 0.191) disorders.

MAGNETIC RESONANCE IMAGING. MRI reports wereavailable for 81 control and 53 PIP patients. The twogroups did not differ significantly for the distributionof the prevalence of MRI abnormalities across the cat-egories of right and left unilateral, bilateral, or no ab-normality (p � 0.738).

CLINICAL SEIZURES AND SEIZURE ONSETS. The analysisregarding seizure onsets and clinical features of seizureswas conducted on 78 control and 53 PIP patients. Six-teen control patients did not produce seizures duringtheir VEEG study, as did 6 PIP patients who did havewell-documented seizures within 24 hours before ad-mission. The groups did not differ for total numbers ofseizures during the monitoring study (p � 0.182). Sec-ondary generalization was more frequent (p � 0.049)in the PIP group, but clustering was not (p � 0.144).

The two groups do not differ with respect to later-alization of seizure onsets across the categories of rightand left unilateral, bilateral, or undetermined (p �0.911). Extratemporal versus temporal onsets (p �0.036) and the occurrence of one or more seizures forwhich both the laterality and localization could not bedetermined were more common in the PIP group (p �0.001).

COMPARISON OF LOCALIZATION BY VIDEO-ENCEPHALO-

GRAPHIC ONSET WITH LOCALIZATION UTILIZING SURGI-

CAL OUTCOME AND INVASIVE MONITORING. Amongthe 131 patients who produced seizure onsets (see Ta-ble 2), Engel class outcomes for at least 6 months aftersurgical intervention were available for 58 of the 66patients who underwent subsequent surgical resection,invasive monitoring, or both; 33 patients with Engelclass 1, and 25 with Engel class 2, 3, or 4 outcomes.The lobar localizations from the VEEG seizure onsetsand surgical resection were concordant in 25 (76%) ofthe class 1 patients. For patients with non–class 1 out-comes, the localization of surgical resection does notconfirm VEEG onsets, because surgical failure can oc-cur regardless of the lobar correspondence between sei-zure onset and resection. In the 25 Engel class 2, 3, or4 patients, the results of invasive EEG monitoring wereused to confirm localization and were concordant withlocalization by VEEG onsets in 18 (72%). Twenty-four of the 25 concordant localizations in the Engelclass 1 patients were temporal and 1 was extratemporal,

in contrast with the 18 concordant localizations in En-gel class 2, 3, or 4 patients, of which 6 were temporaland 12 were extratemporal (p � 0.0001).

INTERICTAL EPILEPTIFORM ACTIVITY AND SLOWING.

Interictal EA was more commonly bilateral in the PIPgroup (p � 0.017). To determine whether lateralizeddifferences accounted for the significant results for the2 � 4 contingency tables for EA, we performed a 2 �2 exact test to compare the groups on only the later-alized (left and right) cells; results were not significant(p � 0.999). The 2 � 4 contingency table for abnor-mal interictal slow EEG activity was not significant(p � 0.152), but the 2 � 2 contingency combining allslow activity indicated that an absence of slow activitywas significantly more common in control patients(p � 0.045).

Multivariate Analysis

PRINCIPAL COMPONENTS ANALYSIS. A set of nine vari-ables was selected from Tables 1 and 2 that appearedto distinguish PIP from control patients, and relate tothe tentative interpretation that PIP is associated withrelatively broadly and bilaterally distributed epilepto-genic networks and genetic determinants of psychiatricdisorders and seizures. As indicated in Table 3, PCAon this set of 9 variables yielded 4 factors with an eig-envalue greater than 1, which accounted for 63.5% ofthe total variance.

These factors appeared to be readily interpretable,with generally comparably high loadings on the vari-ables that defined the factor interpretations, and withsplit loadings (absolute value �0.4) limited to onevariable. The first factor, which constellated undeter-mined and extratemporal (in contrast with temporal)localization, was interpreted as “Ambiguous/extratem-poral onset.” The second factor, on which family psy-chiatric and seizure histories loaded most highly, wasinterpreted as “Family neuropsychiatric history.” Thethird factor, on which bilateral EA loaded positivelyand the absence of abnormal interictal slow activityloaded negatively, was interpreted as “Abnormal inter-ictal EEG.” Lastly, the fourth factor was termed “En-cephalitis” because of the high loading of that variable.

LOGISTIC REGRESSION. Table 4 presents the results ofbivariate logistic regression of PIP on the single vari-able with the highest loading from each of the fourfactors entered separately, compared with the result ofmultiviariate logistic regression on the same four vari-ables together. The similar point estimates of the ORsindicate that the set of four independent variables con-tributed independently to the probability of PIP (Pr[PIP]), with similar effect sizes in the bivariate com-pared with the multivariate logistic regression analyses.

606 Annals of Neurology Vol 63 No 5 May 2008

These logistic regression results suggest that encephali-tis merits inclusion as an independent risk factor, withan effect on PIP that is not solely mediated by seizureonset localization or interictal EEG. The results alsoprovide some support for assigning equal magnitudesand simple additive weights to dichotomous indepen-dent variables, and together with the interpretability ofthe factors and relative strength and specificity of itemloadings suggested an alternative possibility of calculat-ing of Pr (PIP) utilizing a scale composed of the ninevariables entered into the PCA in Table 3.

POSTICTAL PSYCHOSIS RISK SCALE. A simple additivePIP risk scale was computed by assigning the value of�1 to “Temporal seizure onset” and “No interictalEEG slowing,” and the value of �1 to each of theother seven variables in Table 3. Summing these values

by addition yields scores ranging from �2 to 7. Theformula for computing Pr (PIP) using the scale score(S) is as follows:

Pr (PIP) �exp(�1.16 � 0.54S)

1 � exp(�1.16 � 0.54S)

The formula for computation of Pr (PIP) from multi-variate logistic regression on factor scores (F) in Table3 is as follows:

Pr (PIP) �

exp(0.45 � 0.79F1 � 0.61F2 � 0.37F3 � 0.29F4)

1 � exp(0.45 � 0.79F1 � 0.61F2 � 0.37F3 � 0.29F4)

Table 3. Principal Components Analysis with Varimax Rotation on Nine Selected Variables

Variables Factor 1:Ambiguous/Extratemporal

Onset

Factor 2: FamilyNeuropsychiatric

History

Factor 3:Abnormal

Interictal EEG

Factor 4:Encephalitis

Family history of any psychiatric disorder 0.27 0.74a 0.22 0.02

Family history of seizures 0.03 0.60a �0.10 �0.05

History of encephalitis �0.04 0.35 0.07 0.79a

Secondary generalization 0.28 0.46a �0.13 �0.38

Extratemporal seizure onset 0.83a �0.17 �0.22 0.15

Temporal seizure onset �0.85a 0.20 0.20 �0.04

Seizures with undetermined onset 0.67a 0.03 �0.09 0.13

Bilateral interictal epileptiform activity 0.41a 0.02 0.60a �0.40a

No interictal EEG slowing �0.19 0.21 �0.75a �0.24

Variance explained 24.8%a 15.0%a 12.2%a 11.5%a

aFactor loadings with absolute value �0.40.EEG � electroencephalogram.

Table 4. Postictal Psychosis Was Regressed on the Four Independent Variables Loading Most Highly on Each ofthe Four Factors Respectively, Individually in Four Separate Bivariate Analyses (Left), and on All Four VariablesTogether as a Single Multivariate Analysis (Right)

Variable Bivariate Logistic Regression Multivariate Logistic Regression

OR 95% CI p OR 95% CI p

1. Temporal seizure onset 0.44 0.21–0.90 0.024a 0.40 0.19–0.87 0.020a

2. Family history of anypsychiatric disorder 2.53 1.29–4.95 0.007a 2.61 1.21–5.65 0.015a

3. No interictal EEG slowing 0.49 0.25–0.96 0.039a 0.52 0.24–1.13 0.0994. Encephalitis 4.59 1.37–15.41 0.014a 6.62 1.29–33.98 0.024a

The similarity of the point estimates of the odds ratios (ORs) in bivariate and multivariate logistic regression indicates the relativeindependence and comparable effect size of the factors. The result also provides some support for assigning equal magnitudes and simpleadditive weights to dichotomous independent variables in the construction of a postictal psychosis (PIP) risk scale.ap � 0.05.CI � confidence interval; EEG � electroencephalographic.

Alper et al: PIP in Partial Epilepsy 607

Pr (PIP) computed from bivariate logistic regression onthe scale score was strongly correlated with the value ofPr (PIP) obtained from multivariate logistic regressionon the factor scores (r � 0.919; p � 0.0001).

The OR of Pr (PIP) per unit increase in scale scoreis 1.71; (95% confidence interval, 1.36–2.15; p �0.0001). This indicates that for any two subjects with adifference in scale score of �S, the odds of PIP aremultiplied by 1.71�S in the subject with the higherscore relative to the subject with the lower score. TheFigure is the bar plot for the distribution of scale scoresfor control and PIP patients. The Figure depicts eightscale score values between �2 and �5; no subject hada score greater than 5. As is apparent from inspectionof the plot, the control and PIP groups separatestrongly, which is confirmed by the exact test of the2 � 8 contingency table (p � 0.0001). The meanscore in control patients was 0.64 � 1.68 versus2.32 � 1.75 in PIP patients (t � 5.53; degrees of free-dom � 129; p � 0.0001).

RELATION OF SCALE SCORES AND POSTICTAL PSYCHOSIS

TO SURGICAL OUTCOME. Although not a primary aimof the study, bivariate logistic regression analysis indi-cated an association between PIP and Engel class 2, 3,or 4 outcomes (OR, 2.97; 95% confidence interval,1.08–8.18; p � 0.035). However, there was no asso-ciation between PIP and surgical outcome when con-trolling for PIP risk scale score. When surgical out-come was regressed on PIP, score, and theirinteraction, only score had a main effect (p � 0.011),whereas PIP did not (p � 0.909), and there was nosignificant interaction of score with PIP (p � 0.897).

DiscussionThis study found four domains of risk for PIP: ambig-uous/extratemporal localization, family neuropsychiat-ric history, abnormal interictal EEG, and a history ofencephalitis. PIP was associated with bilateral interictalEA, extratemporal or ambiguous seizure onsets, andsecondary generalization. Temporal localization and anabsence of slow activity were more prevalent in controlpatients. PIP was associated with family histories ofpsychiatric disorders and seizures, which themselveswere significantly intercorrelated.

Prior studies have also correlated PIP with bilateralinterictal EA and secondary generalization.1,7,18 Theassociation of extratemporal localization and PIP alsosuggests widely distributed cortical networks19 in viewof the generally larger size of extratemporal versus tem-poral epileptogenic zones.20 A history of encephalitis,which was more prevalent in the PIP group, is report-edly associated with bilaterally and multiply distributedepileptogenic regions.21 However, the PCA and logisticregression results indicate a significant pathogenetic ef-fect of encephalitis that is independent of localizationof seizure onsets or interictal EEG.

In this study, PIP was associated with a greater prev-alence of extratemporal versus temporal seizure onsets.PIP has been associated with temporal lobe epilep-sy7,18,22,23 and has been hypothesized to result fromtemporal limbic dysfunction. However, partial epilepsymay mediate the association with psychosis, rather thanlocalization to the temporal lobe per se. Psychosis isreportedly more frequent in patients with partial thanidiopathic generalized epilepsy,24–26 and the apparentassociation with the temporal lobe may simply reflectthe generally greater prevalence of temporal versus ex-tratemporal onsets in partial epilepsy.27 Extratemporalfoci can involve the temporal lobe in the epileptogeniczone or the region of early spread, which may result inthe misattribution of localization.

There is a significant literature on the relation of lat-erality to psychopathology. Some evidence suggests aparticular role of the left hemisphere in the develop-ment of psychosis.28 No lateralized EEG asymmetrieswere found in this study, consistent with prior studies

Fig. Distribution of scores on the postictal psychosis (PIP) riskscale for control (n � 78; gray bars) and PIP (n � 53;black bars) patients. The score is computed by adding valuesfor each of the nine dichotomous independent variables inTable 3. The value �1 is assigned to temporal seizure onsetand no interictal electroencephalographic (EEG) slowing, andthe value of �1 to each of the other seven variables. No pa-tient in the study scored above 5. The vertical axis representsthe respective proportions of individuals with a given score inthe control and PIP groups, and the absolute numbers areindicated at the top of each bar. The mean score of the con-trol patients was 0.64 � 1.68 versus 2.32 � 1.75 in thePIP group (t � 5.53; degrees of freedom � 129; p �0.0001).

608 Annals of Neurology Vol 63 No 5 May 2008

that also found no association of right or left lateral-ization of seizure onset and PIP.4,7 The more consis-tent underlying factor for PIP risk appears to be abroad distribution of a neural dysfunction, rather thanlaterality.

MRI abnormality did not differ between PIP andcontrol groups with respect to lateralization or overallprevalence. However, this study did not include volu-metric analysis, which could provide substantiallygreater sensitivity and merits investigation in futurestudies of PIP.

PIP in this study was associated with both a familyhistory of seizures and a family history of psychiatricdisorders, which, in turn, were associated with one an-other. The association of a family history of seizureswith a family history of mood and anxiety disorders,but not alcohol use or psychotic disorders, suggests aspecific relation of depression to seizures. Evidence thatsuggests a relation between mood disorders and epilep-sy29 includes the observation of an association of afamily history of mood disorder and PIP,3 epidemio-logical community-based studies that assessed depres-sion retrospectively in incident cases of unprovoked sei-zures and found an association of depression andseizures,30–32 and an apparent anticonvulsant effect ofserotonin reuptake inhibitor antidepressants.33,34

In this study, the relation of family psychiatric his-tory to PIP spanned multiple psychiatric diagnosticcategories and is not restricted only to depression. Thestronger association of PIP to a family history of anypsychiatric disorder than to any of the four underlyingpsychiatric diagnostic categories indicates a more gen-eral relation to psychiatric disorder rather than to anyspecific diagnostic category. This result parallels that ofa recent meta-analysis of psychopharmacological clini-cal trial data that found that the incidence of seizuresin patients assigned to placebo was greater than thepublished incidence of unprovoked seizures in commu-nity nonpatient samples not only in depression, but inpsychotic and obsessive compulsive disorders as well.33

There was a predominance of left-sided cases amongboth PIP and control patients. This may partly reflecta selection bias in the setting of a large surgical epilepsycenter, where cases requiring invasive monitoring andlanguage mapping are often referred. The statisticalanalysis addressed this potential bias by comparing PIPand control groups using contingency tables of eightcells to evaluate the distribution of findings among thecontrol and PIP groups, spanning left lateralized, rightlateralized, bilateral/generalized, and none/not deter-mined. This study found no significant difference be-tween control and PIP groups for the 2 � 4 table thatsubsumed the laterality of seizure onsets, and when thetwo groups were compared with 2 � 2 tables on onlythe subset of lateralized (left and right) cells for inter-ictal EA and for slowing, no differences were seen. The

tendency toward left-sided cases affects both the PIPand control groups, and appears not to have artifactedthe statistical distinction of PIP from control.

The retrospective nature of this study is unlikely tohave significantly diminished the validity of the majorfindings. The methodological significance of a retro-spective versus prospective design relates to the under-lying issues of subject selection bias and blinding. Aprospective design minimizes subject selection bias bysystematic adherence to predefined entry criteria.Blinding of independent from dependent variables in aprospective design is accomplished by temporal order-ing so that independent variables are assessed before,and therefore without knowledge of, the dependentvariables. However, subject selection bias and blindingare the critical issues and not a prospective versus ret-rospective design per se. In this study, selection biasand blinding were addressed by the consecutive natureof the retrospective patient series and by the systematicapplication of the study entry criteria in the selectionof PIP and control cases without knowledge of theVEEG study results.

The additive PIP risk scale developed in this studymay provide a useful independent variable for futureprospective studies. PIP and outcome were associated,but not when accounting for score. Parsing out the un-derlying scale item content that might uniquely andindependently relate to PIP and surgical outcome, re-spectively, will require further analysis on a larger sam-ple.

This study did not utilize a correction for the mul-tiple statistical comparisons. Notwithstanding questionsregarding the validity and usefulness of such correc-tions,35,36 multivariate analysis indicated that the dis-tinction of PIP from control patients could be sub-sumed within a smaller number of underlyingdimensions, providing some assurance regarding thefundamental concern of statistical power. This studyshould be viewed as generating hypotheses that requireindependent confirmation.

In conclusion, ambiguous/extratemporal localization,family neuropsychiatric history, abnormal interictalEEG, and a history of encephalitis were found to beindependent dimensions of risk for PIP. PIP in partialepilepsy appears to be associated with relatively broadlyand bilaterally distributed epileptogenic networks, ge-netic determinants of psychiatric disorders and seizures,and encephalitis.

This research was supported by Finding a Cure for Epilepsy andSeizures (FACES) grant 2006-3 (K.A. and R.K.).

We acknowledge the effort and support of the nursing staff and theelectroencephalography laboratory at the New York UniversityComprehensive Epilepsy Center.

Alper et al: PIP in Partial Epilepsy 609

References1. Logsdail SJ, Toone BK. Post-ictal psychoses. A clinical and phe-

nomenological description. Br J Psychiatry 1988;152:246–252.2. Kanner AM, Stagno S, Kotagal P, Morris HH. Postictal psy-

chiatric events during prolonged video-electroencephalographicmonitoring studies. Arch Neurol 1996;53:258–263.

3. Alper K, Devinsky O, Westbrook L, et al. Premorbid psychiat-ric risk factors for postictal psychosis. J Neuropsychiatry ClinNeurosci 2001;13:492–499.

4. Schmitz B, Wolf P. Psychosis in epilepsy: frequency and riskfactors. J Epilepsy 1995;8:295–305.

5. Tarulli A, Devinsky O, Alper K. Progression of postictal to in-terictal psychosis. Epilepsia 2001;42:1468–1471.

6. Kanemoto K, Kawasaki J, Mori E. Violence and epilepsy: aclose relation between violence and postictal psychosis. Epilep-sia 1999;40:107–109.

7. Umbricht D, Degreef G, Barr WB, et al. Postictal and chronicpsychoses in patients with temporal lobe epilepsy. Am J Psychi-atry 1995;152:224–231.

8. Devinsky O, Abramson H, Alper K, et al. Postictal psychosis: acase control series of 20 patients and 150 controls. Epilepsy Res1995;20:247–253.

9. Savard G, Andermann F, Olivier A, Remillard GM. Postictalpsychosis after partial complex seizures: a multiple case study.Epilepsia 1991;32:225–231.

10. American Psychiatric Association. Diagnostic and statisticalmanual of mental disorders. 4th ed. Washington, DC: Ameri-can Psychiatric Association, 1994.

11. Andreasen NC, Endicott J, Spitzer RL, Winokur G. The familyhistory method using diagnostic criteria. Reliability and validity.Arch Gen Psychiatry 1977;34:1229–1235.

12. Feighner JP, Robins E, Guze SB, et al. Diagnostic criteria foruse in psychiatric research. Arch Gen Psychiatry 1972;26:57–63.

13. Zimmerman M, Coryell W, Pfohl B, Stangl D. The reliabilityof the family history method for psychiatric diagnoses. ArchGen Psychiatry 1988;45:320–322.

14. Duggan C, Sham P, Minne C, et al. Does the method of datacollection affect the reporting of depression in the relatives ofdepressed probands? J Affect Disord 1998;47:151–158.

15. Haut SR, Lipton RB, LeValley AJ, et al. Identifying seizureclusters in patients with epilepsy. Neurology 2005;65:1313–1315.

16. Mehta CR, Patel NR. A network algorithm for performingFisher’s exact test in r � c contingency tables. J Am Stat Assoc1983;78:427–434.

17. Kim J. Factor analysis: statistical methods and practical issues.Thousand Oaks, CA: Sage Publications, 1978.

18. Tebartz Van Elst L, Baeumer D, Lemieux L, et al. Amygdalapathology in psychosis of epilepsy: a magnetic resonance imag-ing study in patients with temporal lobe epilepsy. Brain 2002;125:140–149.

19. Spencer SS. Neural networks in human epilepsy: evidence ofand implications for treatment. Epilepsia 2002;43:219–227.

20. Kutsy RL. Focal extratemporal epilepsy: clinical features, EEGpatterns, and surgical approach. J Neurol Sci 1999;166:1–15.

21. Trinka E, Dubeau F, Andermann F, et al. Clinical findings,imaging characteristics and outcome in catastrophic post-encephalitic epilepsy. Epileptic Disord 2000;2:153–162.

22. Kanemoto K, Kawasaki J, Kawai I. Postictal psychosis: a com-parison with acute interictal and chronic psychoses. Epilepsia1996;37:551–556.

23. Briellmann RS, Kalnins RM, Hopwood MJ, et al. TLE patientswith postictal psychosis: mesial dysplasia and anterior hip-pocampal preservation. Neurology 2000;55:1027–1030.

24. Sachdev P. Schizophrenia-like psychosis and epilepsy: the statusof the association. Am J Psychiatry 1998;155:325–336.

25. McKenna PJ, Kane JM, Parrish K. Psychotic syndromes in ep-ilepsy. Am J Psychiatry 1985;142:895–904.

26. Dongier S. Statistical study of clinical and electroencephalo-graphic manifestations of 536 psychotic episodes occurring in516 epileptics between clinical seizures. Epilepsia 1959;1:117–142.

27. Cascino GD. Complex partial seizures. Clinical features anddifferential diagnosis. Psychiatr Clin North Am 1992;15:373–382.

28. Toone BK. The psychoses of epilepsy. J R Soc Med 1991;84:457–459.

29. Kanner AM. Epilepsy, suicidal behaviour, and depression: dothey share common pathogenic mechanisms? Lancet Neurol2006;5:107–108.

30. Hesdorffer DC, Hauser WA, Annegers JF, Cascino G. Majordepression is a risk factor for seizures in older adults. Ann Neu-rol 2000;47:246–249.

31. Hesdorffer DC, Hauser WA, Olafsson E, et al. Depression andsuicide attempt as risk factors for incident unprovoked seizures.Ann Neurol 2006;59:35–41.

32. Forsgren L, Nystrom L. An incident case-referent study of ep-ileptic seizures in adults. Epilepsy Res 1990;6:66–81.

33. Alper K, Schwartz KA, Kolts RL, Khan A. Seizure incidence inpsychopharmacological clinical trials: an analysis of Food andDrug Administration (FDA) Summary Basis of Approval Re-ports. Biol Psychiatry 2007;62:345–354.

34. Favale E, Rubino V, Mainardi P, et al. Anticonvulsant effect offluoxetine in humans. Neurology 1995;45:1926–1927.

35. Perneger TV. What’s wrong with Bonferroni adjustments. BMJ1998;316:1236–1238.

36. Rothman K. No adjustments are needed for multiple compar-isons. Epidemiology 1990;1:43–46.

610 Annals of Neurology Vol 63 No 5 May 2008