The study was approved by the International Review Board of the Erasmus Medical Centre (Rotterdam, The Netherlands). All patients provided written informed consent. The study was performed on the Mother-Baby Unit (MBU), a five-bed inpatient unit that specializes in the care of patients with severe psychopathology in the postpartum period, located in the Department of Psychiatry in the Erasmus Medical Centre (Erasmus MC) in Rotterdam, The Netherlands. On the MBU, women are admitted with their babies, who stay in a fully staffed nursery adjoining the unit (Bergink et al. 2011). Every patient admitted to the MBU between May 2005 and December 2016 was screened for study inclusion (N = 315).

We included patients with a diagnosis of first-onset mania or psychosis during the postpartum period, who were aged between 18 and 45 years. ‘Postpartum psychosis’ was operationalized as any of the following DSM-IV diagnoses and requiring the specifier ‘onset postpartum’: manic episode, mixed episode, depressive disorder with psychotic features, psychotic disorder not otherwise specified (NOS) or brief psychotic disorder, as assessed with the Structured Clinical Interview for DSM (SCID). Patients were excluded if they had a chronic psychotic disorder, mania, or psychosis with onset during pregnancy or > 12 weeks postpartum, a history of psychosis or mania outside the postpartum period, or drug abuse.

A total of 315 women were admitted to the MBU between May 2005 and December 2016. One hundred thirty-seven of these patients received a diagnosis of postpartum psychosis. Of these, 14 women had a prior postpartum psychiatric episode but no episodes of mania or psychosis at other times. Of the 137 women, four patients declined participation. In addition, 21 women were excluded: 18 women were excluded because they had a history of mania or psychosis outside the postpartum period, one woman was excluded because of postpartum drug abuse, one woman was excluded because her symptom onset was > 12 weeks postpartum, one woman was excluded because her symptoms started during pregnancy. Accordingly, 112 patients fulfilled the criteria for first-onset postpartum psychosis. Five patients were lost to follow-up (4.5%) and one patient (0.9%) was lost to suicide (baseline and clinical characteristics of these women can be found in Additional file 1: Table A1). In this study, we therefore included 106 women admitted to the MBU between 2005 and 2016. None of the women were breastfeeding during their MBU admission.

Patients were diagnosed by a clinician using the Structured Clinical Interview for DSM (SCID-1/P research version) (First et al. 1999). The SCID is a semi-structured interview guide for making diagnoses according to the diagnostic criteria published in the American Psychiatric Association’s Diagnostic and Statistical Manual for Mental Disorders (DSM). Previous hypomanic and manic episodes were also registered using the SCID. We further assessed demographics, psychiatric history, and family history of psychiatric illness (Table 1) (for more detail, see 12).

Phenomenology of the initial episode was assessed using the Bipolar Affective Disorder Dimension Scale (BADDS) (Craddock et al. 2004). The BADDS comprises four dimensions which provide a quantitative measure of psychopathology in each of four domains: (1) Manic-like episodes (the Mania dimension, M), (2) Depression-like episodes (the Depression dimension, D), (3) Psychotic symptomatology (the Psychosis dimension, P) and (4) the relationship (congruence of content and timing) between psychotic features (if present) and mood episodes (the Incongruence dimension, I). Each dimension provides a composite measure that takes both severity and frequency of relevant psychopathology into account. The dimensions are rated using integers in the range 0–100, with higher scores indicating more clinically important psychopathology—typically a mix of severity and frequency/duration.

Because clinical presentation, family history, and the longitudinal illness course overlap markedly with those of bipolar disorder, postpartum psychosis is generally considered a bipolar spectrum illness and not a primary psychotic disorder (Sit et al. 2006; Chaudron and Pies 2003; Jones and Craddock 2002). In the absence of formal guidelines, treatment in clinical practice is typically based on the most prominent symptom dimensions. Benzodiazepines are used for insomnia and agitation, antipsychotics and mood stabilizers for psychotic and manic symptoms, and antidepressants for depressive symptoms.

During admission, women with a first-onset postpartum psychosis were treated according to a standardized treatment algorithm, as described previously (Bergink et al. 2015b). The first step in treatment involves benzodiazepines at bedtime for three days (step 1). The purpose of starting with an initial period of benzodiazepine monotherapy is to evaluate whether restoration of sleep results in clinical remission of manic and psychotic symptoms, as sleep loss has been considered an important etiological factor in postpartum psychosis (Sharma and Mazmanian 2003). For patients whose manic or psychotic symptoms persist after a few days of benzodiazepine monotherapy, the next recommended step involves antipsychotic medication (step 2). Although the efficacy of antipsychotics in the absence of mood stabilizers has been described in only three case reports, antipsychotics are frequently used worldwide as first-line treatment for patients with postpartum psychosis and mania (Doucet et al. 2011). Furthermore, antipsychotics are often considered the preferred pharmacological treatment option for acute mania outside the postpartum period (Cipriani et al. 2011). Our primary recommendation for antipsychotic treatment was haloperidol at 2–6 mg/day. Patients who experienced side effects were switched to an atypical antipsychotic. A subset of patients who had already been treated with an antipsychotic before admission (e.g., by acute services) were continued on the same antipsychotic they received before admission. If there was no significant clinical response after 2 weeks, adjunctive lithium was initiated (step 3). One small open-label study and a case report have suggested that the combination of lithium and antipsychotics is more effective than antipsychotic monotherapy in patients with postpartum psychosis (Targum et al. 1979; Silbermann et al. 1975). Furthermore, studies have demonstrated efficacy of lithium in the prevention of postpartum psychosis (Stewart et al. 1991; Cohen et al. 1995; Austin 1992; Bergink et al. 2012). Lithium dosing was achieved based on plasma level (target, 0.8–1.0 mmol/L).

After complete remission of symptoms, all women were advised to taper benzodiazepines to discontinuation. Women receiving antipsychotic monotherapy were advised to continue this treatment as maintenance therapy until nine months postpartum. Women who achieved clinical remission using both antipsychotics and lithium were advised to gradually taper off antipsychotic treatment, with maintenance lithium monotherapy until nine months postpartum. Lithium dosing for relapse prevention was achieved based on plasma level (target, 0.6–0.8 mmol/L).

Four years postpartum, women were re-evaluated using the SCID (First et al. 1999). Women were not seen in between hospital discharge and follow-up for the purposes of this study. Recurrence was defined as the occurrence of any depression, (hypo)mania, psychosis or mixed state episode fulfilling DSM-IV criteria, admission to hospital or a restart of medication. All women with a recurrence were asked retrospectively about the timing of their episode, including whether this was in relation to a subsequent pregnancy. Additionally, we collected information on the timing of tapering or stopping medication if applicable. The patient’s medical records were consulted to validate the information.

Based on information collected at follow-up, women were categorized into one of two groups: (1) women with recurrence of non-postpartum mood or psychotic episodes within the follow-up period, or (2) women with mania/psychosis in the postpartum period and no mood or psychotic episodes outside the postpartum period during follow-up (vulnerability to affective psychosis only after childbirth).

We summarize the longitudinal course of the illness per in Table 1. Differences between the two groups in terms of baseline demographic and clinical characteristics were assessed using Chi-squared and t-tests were appropriate (Table 1). A Kaplan–Meier Survival Curve of recurrence rates within the four-year follow-up period after first-onset postpartum psychosis was plotted. Additionally, we used a binomial logistic regression model to identify clinical predictors of postpartum psychosis group (recurrence of non-postpartum mood or psychotic episodes vs. mania/psychosis in the postpartum period only). Potential predictors of recurrence were based on the literature included admission length, maternal age, phenomenology of the index episode, and family history of psychiatric illness (Kapfhammer et al. 2014; Blackmore et al. 2013; Benvenuti et al. 1992). To improve power in the family history variable, depression and anxiety were combined into the category ‘depression or anxiety’ and postpartum depression and postpartum psychosis were combined into the category ‘postpartum psychiatric episode’. Mania/psychosis in the postpartum period only and psychiatric disorder with non-postpartum episodes were coded as 0 and 1 respectively. Results are presented in the form of odds ratios. All statistical analyses were performed using Stata/MP 15 (StataCorp 2017). Lastly, we explored whether the set of psychotic symptoms recognized as having special weight in the diagnosis of schizophrenia and schizoaffective disorder (thought echo, insertion, withdrawal, or broadcasting; passivity experiences; hallucinatory voices giving running commentary, discussing subject in third person, or originating in some part of the body; bizarre delusions; catatonia) was a precursor for a diagnosis within the psychotic disorder spectrum.