Association of pharmacological treatments and real‐world outcomes in borderline personality disorder

Most patients with borderline personality disorder (BPD) receive psychopharmacological treatment, but clinical guidelines on BPD lack consensus on the role of pharmacotherapy. We investigated the comparative effectiveness of pharmacological treatments for BPD.

among individuals with BPD. No such associations were found for benzodiazepines, antidepressants, antipsychotics, or mood stabilizers.

K E Y W O R D S
borderline personality disorder, pharmacoepidemiology, real-world data 1 | INTRODUCTION Borderline personality disorder (BPD), which affects about 1%-2% of the general population, 1,2 has no established pharmaceutical treatment. Nevertheless, up to 94% of patients with BPD receive psychotropic medication. 3,4 Aims to treat psychiatric comorbidities (e.g., depression) frequently lead to pharmacological treatment in BPD. Also, in the absence of available psychotherapeutic treatments, pharmacotherapy may be preferred. 5 Attempts to address the symptoms of BPD with pharmacotherapy during a series of individual crises may eventually result in the accumulation of polypharmacy in patients with BPD.
Clinical guidelines lack clear consensus on the role of pharmacotherapy in treating BPD. In particular, while some recommendations expressly advise against using pharmacotherapy, other guidelines view medication as adjunctive therapy in the treatment of issues like impulsivity, cognitive-perceptual symptoms, or anger. [6][7][8] However, the efficacy of pharmacotherapeutic treatments for BPD is unclear.
Recent meta-analysis pooling prior randomized controlled trials (RCTs) on pharmacotherapies for BPD (21 studies, N = 1768 patients) indicated that treatment with antidepressants, mood stabilizers, or antipsychotics was not associated with symptom relief for BPD. 9 Similarly, the most recent Cochrane review found that treatment with antidepressants, antipsychotics, or mood stabilizers had little to no effect on BPD symptoms. 10 Furthermore, although over half of the patients are treated with benzodiazepines, 3 there have never been any placebo-controlled double-blind studies on benzodiazepine treatment for BPD. Finally, despite 38% comorbidity for ADHD in patients with BPD, 11 the use of ADHD medications in the treatment of BPD has received less attention.
Using RCTs' results to evaluate pharmacotherapies' usefulness in BPD treatment presents a number of impediments. First, most patients with BPD would be excluded from a typical RCT owing to prevalent comorbidities, such as substance abuse or suicidal behavior, 12,13 thereby compromising the generalizability to real-world patients. Second, conventional RCTs concentrate on reducing specific BPD symptoms over a brief follow-up period. 14 Consequently, the long-term effects of pharmacological treatment in BPD (e.g., risk of hospitalization or death) have remained unknown.
Observational studies can overcome some of the aforementioned issues by leveraging large, unselected nationwide electronic databases with long follow-ups. A major challenge in observational studies is selection bias since the treatments are not randomized. However, this can be overcome by employing within-individual modeling in which each patient serves as his or her own control. 14 To our knowledge, only a handful of small studies with a few different treatment arms have compared the effectiveness of pharmacological treatments for BPD, 15,16 and no large-scale observational investigation on this topic exists. Here we aimed to study the comparative effectiveness of commonly used pharmacological treatments of BPD in an unselected Swedish nationwide cohort.

Significant outcomes
• Compared with individuals' non-use periods, ADHD medications were the only pharmacological group associated with reduced risk of psychiatric rehospitalization or hospitalization owing to any cause or death among individuals with borderline personality disorder. • Treatment with benzodiazepines, antidepressants, antipsychotics, or mood stabilizers did not associate with a reduced risk of psychiatric rehospitalization or hospitalization owing to any cause or death.

Limitations
• We used real-world data from nationwide register databases that lacked specific clinical parameters (e.g., the severity of BPD symptoms, pharmacological indications, quality of life, and level of functioning) and concomitant psychotherapeutic treatments. • Protopathic bias might have affected our findings, although attempts were made to mitigate its potential effect.  December 31, 2018. The followup started at the first diagnosis or January 1, 2006, for those diagnosed before that. Individuals with comorbid non-affective psychotic disorders (ICD-10: F20-F29), bipolar disorder (ICD-10: F31), psychotic depression (ICD-10: F32.3), and personality disorders other than BPD were excluded (ICD-10: F60-69 except F60.3). The analyses were also censored to these diagnoses, in addition to death, emigration, and end of data linkage (December 2018).

| Exposure
We investigated medication exposure at both the group (e.g., antipsychotic treatment) and individual agent levels (e.g., treatment with escitalopram). The reference for each group and the specific drug was non-use of that medication group (e.g., escitalopram was compared with non-use of antidepressants). For measuring medication exposure, we employed the ATC classification recommended by the World Health Organization (https:// www.whocc.no/atc/structure_and_principles/). Medication classifications, according to ATC, are subdivided into 14 main categories (Level 1) and up to four levels. We classified treatment exposure as follows: antipsychotics (ATC: N05A), antidepressants (N06A), mood stabilizers (N03AF, N03AG, N03AX, N05AN01), benzodiazepines and related pharmaceuticals (N05BA, N05CD, N05CF), and ADHD medications (N06BA). We created the drug use periods based on prescription drug purchases utilizing the PRE2DUP approach, which is described in detail elsewhere. 17 Briefly, the PRE2DUP approach is based on the calculation of sliding averages of defined daily dosages, the quantities of medications purchased, and individual drug use patterns. The PRE2DUP modeling also incorporates hospital stays and medicine stockpiling. Medications having fewer than 100 patients or fewer than 50 outcomes were not reported. However, we presented results for clozapine (30 patients) and long-acting injectables (LAIs) (47 patients) owing to prior research interest 16,18,19 and demonstrated high effectiveness in schizophrenia. 20

| Outcomes
Our main outcomes were (1) any psychiatric rehospitalization (ICD-10: diagnoses beginning with the letter F), which served as a surrogate for treatment failure, and (2) hospitalization owing to any reason or death, which served as a proxy for both treatment failure and tolerability of treatments. To control protopathic bias, 21 we ran sensitivity analyses in which the first month of medication and non-medication exposure were excluded from the follow-up. 14 Specifically, pharmacological treatments in BPD are often initiated during sudden crises, which may bias the findings toward poor outcomes owing to the time required for pharmacotherapies to reach full efficacy. As supplementary analyses, we also investigated all-cause mortality and work disability (i.e., the start of sickness absence of over 14 days or being granted a disability pension).

| Statistical analyses
For statistical analysis, we employed SAS version 9.4. We modeled our primary outcomes as recurrent events. Next, we analyzed them using the within-individual Cox regression model in which each individual forms his or her own stratum, thereby eliminating selection bias (i.e., non-users may differ from users of a particular pharmacotherapy). By using within-individual analysis, the effect of exposure on the outcome is estimated while controlling for all time-invariant confounding factors, such as genetics and childhood environment. In withinindividual design, the model needs to be only adjusted for time-varying factors such as time since onset of illness, the temporal order of treatments, and other concomitant pharmacotherapy. This minimizes the risk of selection bias, as the study design does not depend on the selection of a specific group of participants but rather uses all available data from each participant. The method is detailed elsewhere. 22 We adjusted our models for the temporal order of treatments, time since cohort entry, and the use of psychotropic drugs, including antidepressants, benzodiazepines, and related pharmaceuticals, mood stabilizers, ADHD medications, and medications used to treat substance use disorders (disulfiram, acamprosate, naltrexone, nalmefene, buprenorphine, methadone). We presented the outcomes using hazard ratios (HR) with confidence intervals of 95%. (CIs). We applied the Benjamini-Hochberg False Discovery Rate (FDR) correction in each outcome analysis to control false positives owing to multiple comparisons. 23 In the within-individual analysis, the follow-up time is reset to zero after each outcome event so that treatment durations within a person can be compared. 14 We also conducted traditional between-individual analyses to compare the results to the above-mentioned within-individual analyses and to analyze mortality (onetime event). Specifically, for between-individual analyses, we deployed multivariate-adjusted Cox regression models, adjusted for sex, age, educational level, the number of previous hospitalizations owing to BPD, time from first BPD diagnosis, comorbidities, and other medications. Compared with within-individual analyses where individuals are compared with themselves, all individuals contribute to the analysis in between-individual analyses.

| RESULTS
As shown in Table 1, we collected data on 17,532 BPD patients with an average baseline age of 29.8 years who T A B L E 1 Characteristics of the study population (N = 17,532).
The relationships between specific pharmacotherapies and the two primary outcomes are depicted in outcomes. Of all the explored antidepressant medications, we discovered that paroxetine was associated with the highest risk of psychiatric rehospitalization (HR = 1.36, 95% CI = 1.15-1.60) and the highest risk of all-cause hospitalization or death (HR = 1.32, 95% CI = 1.14-1.52). Of the investigated antipsychotics, the highest risk of psychiatric rehospitalization was observed for treatment with haloperidol (HR = 1.49, 95% CI = 1.15-1.92), whereas zuclopenthixol was associated with the highest risk of all-cause hospitalization or death (HR = 1.66, 95% CI = 1.32-2.09). The above within-individual analyses aligned with traditional between-individual analyses (Supplementary Tables 3  and 4 and Supplementary Figure 1). Specifically, the rank F I G U R E 2 The association between use versus non-use of individual pharmacotherapeutic agents and risk of psychiatric rehospitalization. The asterisk presents an FDR-corrected p-value of <0.05. order of hazard ratios from within-individual analyses was comparable to that of hazard ratios from between-individual analyses for both main outcomes: psychiatric hospitalization (Spearman's rho = 0.72) and all-cause hospitalization or death (Spearman's rho = 0.82).

| DISCUSSION
To our knowledge, this is the first study on the comparative effectiveness of psychopharmacological treatments for BPD. Our main findings indicate that several commonly prescribed medications to treat BPD are associated with poor effectiveness or even adverse outcomes, such as a significant risk of psychiatric rehospitalization, even when potential protopathic bias was controlled. To address potential protopathic bias, we conducted sensitivity analyses (i.e., 1 month of censoring from the beginning of each drug use period), and the rank order of medications was equal to that of the primary analyses, demonstrating clinically significant differences among F I G U R E 3 The association between use versus non-use of individual pharmacotherapeutic agents and risk of all-cause hospitalization or death. The asterisk presents an FDR-corrected p-value of <0.05. the pharmacological treatments used. Note that even in the presence of possible protopathic bias, treatment with lisdexamphetamine, bupropion, methylphenidate, and clozapine was associated with improved outcomes, encouraging further research on these treatments. We are unaware of any thresholds for clinically significant hazard ratios in psychiatry. Intriguingly, the rigorous hazard ratio criterion (i.e., HR <0.8) utilized in oncology would deem the relationships between treatment with clozapine or lisdexamphetamine and psychiatric rehospitalization to be clinically significant. 24 The majority of patients with BPD are treated with psychotropic medications, and considerable polypharmacy is common. 3,25,26 Based on our findings, some of the medications recommended by clinical guidelines (e.g., olanzapine, quetiapine, and selective serotonin reuptake inhibitors [SSRIs]) are potentially harmful, while other recommended treatments (e.g., lamotrigine and valproate) appear to be ineffective. 6,8 The disparity between these recommendations and our findings may be owing to the fact that the clinical guidelines are primarily based on a small number of RCTs with highly selected samples and short follow-up periods. In fact, a prior systematic review and meta-analysis revealed that just four previous RCTs on BPD treatment had more than 70 participants. 9 There has been a decline in the number of pharmacological publications conducted on individuals with BPD, which may suggest pessimism over the efficacy of pharmacotherapy in BPD. In fact, just a handful of RCTs have been published in recent years. 19,27,28 However, there is ongoing industry-funded research on novel pharmacotherapeutic treatments for BPD. 29 In accordance with prior epidemiological research, 3,25 more than half of the patients were administered benzodiazepines and related drugs. Typically, benzodiazepines are used to alleviate abrupt distress or anxiety during times of crisis. Nevertheless, given previous pharmacoepidemiologic research indicating that benzodiazepine treatment lasting more than 4 weeks is common, 30 there is a considerable risk of short-term drug treatment becoming long-term treatment in BPD patients. A previous cross-over trial demonstrated that benzodiazepine treatment is associated with amplified impulsivity in patients with BPD. 31 In this light, it is plausible that our findings on benzodiazepine treatment-related poor outcomes are attributable to an increase in impulsivity, which is a risk factor for, for example, suicidal behavior. 32 We also examined the association between clozapine and LAI antipsychotic treatment and real-world outcomes in BPD. We found that treatment with clozapine (but not with LAI antipsychotics) was associated with a reduced probability of psychiatric readmission. Note, however, that patients receiving clozapine treatment were rare in our cohort. The average dose of clozapine administered to individuals with BPD was lower than the average dose administered to patients with schizophrenia (229 vs. 426 mg/d). 33 It is possible that some of these patients will be diagnosed with a psychotic disorder in the future, given that the primary indication for clozapine is treatment-resistant schizophrenia. 34 Nevertheless, our findings align with the previous Danish nationwide observational study that also found an association between treatment with clozapine and a reduction in psychiatric admissions in patients with BPD. 18 Altogether, to date, as the only RCT on clozapine failed to enroll a significant number of patients with BPD, 19 the evidence supporting the treatment of clozapine in BPD is primarily based on a few observational studies and single-patient cases. 35 We discovered that treatment with pharmacological agents that raise extraneuronal dopamine and norepinephrine concentrations (i.e., bupropion, lisdexamphetamine, and methylphenidate) 36,37 was associated with improved real-world outcomes in BPD. Owing to the fact that these medications are widely used to treat patients with ADHD, it is possible that patients receiving them also exhibit ADHD symptoms. Although BPD and ADHD partially overlap in symptoms such as impulsivity and emotion dysregulation, 38,39 previous efforts to investigate the efficacy of ADHD medication treatment in BPD are scarce. One small previous RCT found that administration of methylphenidate was associated with improvement in decision-making in patients with BPD with low inattention problems. 40 Two open-label studies found that methylphenidate treatment in BPD was associated with reduced aggression, symptom severity, and impulsiveness. 41,42 Our results indicate that treating patients with BPD using stimulants is potentially effective. However, future RCTs are required to determine whether such treatments should be administered to patients with BPD without comorbid ADHD symptoms.

| Strengths and limitations
Our study has both strengths and limitations. We collected a large, unselected sample of patients with BPD treated with commonly used pharmacotherapies with up to 12-year follow-up, thereby increasing our findings' generalizability to real-world patients. This is a significant strength of the present study, given that majority of patients with BPD frequently exhibit suicidal behavior and substance use disorder, 12,13 which would exclude them from a standard clinical trial. However, our register databases lacked specific clinical parameters such as the severity of BPD symptoms, pharmacological indications, quality of life, and level of functioning. Also, we lacked information regarding concomitant psychotherapy treatments suggested as the first-line treatment option for individuals with BPD in major clinical guidelines. 6,7 A previous study has shown that dialectical behavioral therapy treatment could help reduce the medication load in BPD. 43 Further, our sensitivity analysis may not have fully eliminated protopathic bias. Therefore, our findings on the associations between the explored pharmacotherapies and, for example, the risk of psychiatric rehospitalization may give a too negative view of the effectiveness of pharmacological treatments. On the other hand, it is also possible that treatment with clozapine, lisdexamphetamine, bupropion, or methylphenidate, in reality, might have a more substantial effect in reducing the risk of psychiatric rehospitalization than the observed associations. Finally, our results may generalize only to high-income countries with a state-funded health care system.
To conclude, our findings from a large, unselected national cohort suggest that many widely used pharmacological treatments for BPD may not associate with a reduced risk of psychiatric hospitalization and hospitalization owing to any cause or death. However, we found support for the use of lisdexamphetamine, bupropion, methylphenidate, and clozapine in BPD treatment, but further research is needed before these treatments should be more widely adopted.

AUTHOR CONTRIBUTIONS
Johannes Lieslehto, Heidi Taipale, and Jari Tiihonen conceptualized the paper. Jari Tiihonen, Markku Lähteenvuo, Heidi Taipale, Ellenor Mittendorfer-Rutz, and Antti Tanskanen oversaw data collection and project development. Heidi Taipale, and Antti Tanskanen were responsible for the statistical analyses. Johannes Lieslehto, Heidi Taipale, and Jari Tiihonen drafted the manuscript and provided data interpretation. Markku Lähteenvuo, and Ellenor Mittendorfer-Rutz assisted with the data interpretation. All authors participated in finalizing the manuscript, agreed upon the final version of the manuscript and meet the definition of an author, as stated by the International Committee of Medical Journal Editors.

FUNDING INFORMATION
This work was supported by the Finnish Ministry of Social Affairs and Health through the developmental fund for Niuvanniemi Hospital, HT by the Academy of Finland (grants 315969, 320107, 345326). The study's funders had no role in the study design, data collection, data analysis, data interpretation, or writing of the report. We utilized data from the REWHARD consortium supported by the Swedish Research Council (grant number 2017-00624).