Pharmacokinetic Markers of Clinical Outcomes in Severe Mental Illness: A Systematic Review

The term severe mental illness (SMI) encompasses those psychiatric disorders exerting the highest clinical burden and socio-economic impact on the affected individuals and their communities. Pharmacogenomic (PGx) approaches hold great promise in personalizing treatment selection and clinical outcomes, possibly reducing the burden of SMI. Here, we sought to review the literature in the field, focusing on PGx testing and particularly on pharmacokinetic markers. We performed a systematic review on PUBMED/Medline, Web of Science, and Scopus. The last search was performed on the 17 September 2022, and further augmented with a comprehensive pearl-growing strategy. In total, 1979 records were screened, and after duplicate removal, 587 unique records were screened by at least 2 independent reviewers. Ultimately, forty-two articles were included in the qualitative analysis, eleven randomized controlled trials and thirty-one nonrandomized studies. The observed lack of standardization in PGx tests, population selection, and tested outcomes limit the overall interpretation of the available evidence. A growing body of evidence suggests that PGx testing might be cost-effective in specific settings and may modestly improve clinical outcomes. More efforts need to be directed toward improving PGx standardization, knowledge for all stakeholders, and clinical practice guidelines for screening recommendations.


Introduction
Mental and substance use disorders are leading causes of disability on a global level [1], with a significant portion of this burden deriving from severe mental illnesses (SMIs) [2]. Collectively, SMI represents an ill-defined category which has been inconsistently reported in the literature in the field [3] but that, as a bare minimum, comprises conditions such as schizophrenia (SCZ), bipolar disorder (BD), and major depressive disorder (MDD) [2,3]. Among individuals affected by SMI, life expectancy has been reported to be reduced by 20 years among males and up to 15 years among females [4]. In the past, this gap in life expectancy was frequently attributed to suicide risk. However, over the years, it has been increasingly evident how cardiovascular and infectious disorders also represent significant causes of death in this population [4][5][6]. The toll associated with SMI is not limited to the affected individuals but extends to their relatives and communities [7]. Carers of individuals affected by SMI may indeed report lower employment levels, and social and economic difficulties with higher levels of food insecurities [8] and expenditures related to care [9]. Individuals affected by SMI represent a severely underserved population, despite significant Int. J. Mol. Sci. 2023, 24, 4776 2 of 20 advancement in their management. For example, only 41% of individuals affected by MDD may receive treatment at minimal standard of care [10]. Even for the minority of individuals receiving treatment, finding the most effective therapeutic option could be challenging for healthcare providers and service users. In fact, even when the most updated protocols are employed, the treatment choice is based on a "trial-and-error" approach, which ultimately may result in frequent treatment failures and significant healthcare costs [11,12]. Numerous factors should be considered when discussing the basic underpinnings for the observed heterogeneity in treatment response (HTR), such as the nosological classification systems used for the diagnoses [13][14][15][16], age of onset, co-morbidities, and clinical course. These factors likely represent a source of HTR intrinsic to the current standards of practice [17]. Notwithstanding the previously mentioned limitations, this framework has produced most of the evidence for treatments (either pharmacological or psychotherapy) in psychiatry since clinical trials testing the efficacy and tolerability of a particular intervention have indeed selected study patients based on a categorical nosological system [17]. Waiting for the development of more accurate diagnostic tools [18], one possible way to address HTR would be to tailor treatments to the individuals identified through the use of the current nosological classification systems by matching the right treatment to the right patient [19][20][21][22]. In this setting, a growing body of evidence suggests that pharmacogenomics (PGx) may represent a useful tool for enabling personalized treatments. PGx is the research area dedicated to evaluating how multiple genetic variations may interact and influence the metabolism and action of a particular pharmacological treatment [23]. With very few notable exceptions (e.g., lithium salts, gabapentin), nearly all medications currently employed for the treatment of psychiatric disorders are metabolized in the liver. The major metabolic reactions involved in the process are oxidation (phase I) and conjugation (phase II). Genetic variations for transporters expressed at different locations, such as the brain, gut, and liver, can also influence the pharmacokinetic profile of the different compounds employed in treatment, but their clinical impact has not been established [23]. The metabolic system that has been most extensively studied is represented by cytochrome P450 (CYP450), comprising 57 genes and 58 pseudogenes [24]. The two isoenzymes of CYP450 most extensively studied for psychiatric treatments are CYP2D6 and CYP2C19, as there is significant evidence that these two can significantly influence psychotropic metabolism [24,25], with CYP2D6 being involved in the metabolism of almost half of the most prescribed psychotropics [25]. For a long period of time, it has been known that single-nucleotide polymorphisms (SNPs) could be associated with differential gene expression profiles and that these, in turn, could be studied to help estimate the risk of developing adverse effects or to quantify treatment response to a particular medication in a subgroup of individuals [22]. Allelic variants of CYP genes are indicated with an asterisk (*), genotypes are then coded based on their projected metabolic activity, and the corresponding phenotypes are typically subdivided into Rapid, Normal, Ultrarapid, Intermediate, and Poor Metabolizer [23]. Genes supposedly associated with the postulated mechanism of action at the biochemical, cellular, and physiological level are instead associated with the pharmacodynamic of a particular compound. In psychiatry, attention has been focused on possible allelic of genes involved in neurotransmitters' receptors, signal transmission, gene transcription, or protein folding, among others [23]. Gene variations in human leukocyte antigens or in proteins regulating immune mechanisms have also been the subject of research in the area and have yielded guidance in the projected risk of developing adverse reaction upon exposure to certain compounds [23,26]. To improve the accessibility to treatment-informing guidance based on PGx, several scientific bodies have developed clinical practice guidelines with the most significant being summarized on easily accessible platforms such as PharmGKB [26]. In theory, PGx holds great promise in terms of improving personalization of treatments as it would aid clinicians in streamlining the pharmacological treatment selection based on the expected efficacy and tolerability for the different available pharmacological treatments [11]. However, in psychiatry the clinical application of this tool has lagged behind due to concerns regarding its efficacy and lack of knowledge on interpreting its results in a sizeable portion of healthcare providers. In the present study, we performed a systematic review of the literature in the field probing the use of PGx for SMI, specifically reporting on pharmacokinetic markers of treatment response, as defined by the authors. Importantly, we applied for the first time a transdiagnostic approach to explore whether we could be able to identify PGx markers associated with similar patterns of response across disorders. The main objective of this project is reviewing the existing evidence for pharmacokinetic markers in predicting pharmacological treatment response in individuals affected by SMI, focusing on the comparison with the usual standard of care when available.

Materials and Methods
A double-blind systematic review was performed on Scopus, PubMed, and Web of Science according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA [27]). In this project, we considered including articles published in English probing the association of PGx tests with pharmacological treatment outcomes for SMI (i.e., BD, MDD, SCZ) and reporting on pharmacokinetic markers. We defined treatment outcomes as a response to the practiced treatment regimen, as reported by the authors. Accepted study designs included: (1) open-label trials, (2) randomized controlled trials, (3) cross-sectional studies, (4) retrospective cohort studies, (5) prospective cohort studies, and (6) studies recruiting human subjects ≥ 18 years old and assessing treatment outcomes as defined by the study authors. We excluded: (1) meta-analyses, (2) systematic reviews, (3) case reports, (4) case series, (5) letters to the editor, and (6) editorials. No time restriction was applied based on the year of publication. Pharmacodynamic markers and studies assessing the safety or tolerability profile of pharmacological treatments were excluded. The following search strategy was employed ("pharmacogenomic" OR "pharmacogenomics" OR "pharmacogenetics" OR "pharmacogenetic") AND ("signature" OR "biomarkers" OR "marker" OR "determinants") AND ("severe mental illness" OR "severe mental disorders" OR "schizophrenia" OR "psychosis" OR "schizoaffective disorder*" OR "bipolar disorder *" OR "major depressive disorder *"). Two reviewers independently screened the records identified through the primary search strategy. With the objective of reviewing the existing evidence for pharmacokinetic markers in predicting pharmacological treatment response in individuals affected by SMI, we focused on extracting the following data from the included studies: (1) study design, (2) sample composition, (3) main objective, (4) inclusion and (5) exclusion criteria, (6) country where the selected study was performed, and (7) reported outcomes pertinent to our project. The qualitative data extraction was performed independently by two authors (P.P.; L.B.) and whenever a discrepancy was found a third senior author was involved to reach a consensus. Rayyan, a semi-automated tool, was employed to facilitate the screening process [28]. The primary search was further augmented using a comprehensive pearl-growing strategy. ROB 2 [29] was employed for the assessment of bias for randomized controlled trials by two independent raters. Again, discrepancies were solved through discussion and, if needed, with a third author's judgement. The last search was performed on the 17 September 2022. All tables are available in interactive mode on GitHub (https://github.com/claudiapis/tables_pharmacokinet ic_markers, accessed on 25 February 2023). Further, the main input set is available on GitHub (https://github.com/pasqualeparibell/Pharmacokinetic-markers-of-clinical-ou tcomes-in-severe-mental-illness-a-systematic-review.---source/tree/main, accessed on 25 February 2023).

Search Results and Bias Assessment
The selected search strategy resulted in the identification of 1975 records. After duplicate removal, 1456 records were assessed through an abstract and title screening, leading, in turn, to the identification of 587 records. Among them, 42 papers were selected for the qualitative analysis, summarized in 3 different tables dedicated to (1) SCZ (Tables 1 and 2)

Search Results and Bias Assessment
The selected search strategy resulted in the identification of 1975 records. After duplicate removal, 1456 records were assessed through an abstract and title screening, leading, in turn, to the identification of 587 records. Among them, 42 papers were selected for the qualitative analysis, summarized in 3 different tables dedicated to (1) SCZ (Tables 1  and 2) MDD (Tables 2 and 3) and BD (Table 3). A complete description of the selection process is reported in the PRISMA flow diagram, in Figure 1.  A total of 13 studies originated in the USA and 14 in Asia. The remaining studies were carried out mainly in European countries. Among the included studies, eleven were randomized controlled trials (RCTs), with 10 recruiting individuals affected by MDD and only one focusing on individuals affected by SCZ [30]. One RCT on MDD recruited a mixed sample of individuals affected by MDD and/or anxiety, but no description of the anxiety disorder was included [31]. Only three of the included studies reported on individuals affected by BD, with one of the three including a heterogeneous population comprising MDD, BD, and post-traumatic stress disorder (PTSD) [32]. Overall, the risk of bias of the included RCTs appears limited, save for three studies, judged at high risk of bias [33][34][35]. Figure 2 summarizes the bias assessment for the included RCTs according to ROB 2. A total of 13 studies originated in the USA and 14 in Asia. The remaining studies were carried out mainly in European countries. Among the included studies, eleven were randomized controlled trials (RCTs), with 10 recruiting individuals affected by MDD and only one focusing on individuals affected by SCZ [30]. One RCT on MDD recruited a mixed sample of individuals affected by MDD and/or anxiety, but no description of the anxiety disorder was included [31]. Only three of the included studies reported on individuals affected by BD, with one of the three including a heterogeneous population comprising MDD, BD, and post-traumatic stress disorder (PTSD) [32]. Overall, the risk of bias of the included RCTs appears limited, save for three studies, judged at high risk of bias [33][34][35]. Figure 2 summarizes the bias assessment for the included RCTs according to ROB 2.

PGx Outcomes
Reported outcomes included service use reduction, symptom change from baseline, and rates of remission or response to treatment. In line with the literature in the field, there was significant heterogeneity in scales employed to report the symptom changes. The description of the sample composition of the included studies also appears inconsistent, with the vast majority providing the gender composition, age range, or the average age of the recruited sample. A discrete heterogeneity also emerged regarding the employed inclusion or exclusion criteria, even considering the heterogeneity of the analyzed diagnostic categories. Numerous different alleles and genotypes have been assessed, but no specific efficacy pattern emerged for a particular marker across the various studies. A relatively limited number of studies [32,34,36,37,39,[41][42][43][44] reported on the results of combinatorial PGx testing, introducing a further layer of complexity in the interpretation of the results for the included studies.

PGx Outcomes
Reported outcomes included service use reduction, symptom change from baseline, and rates of remission or response to treatment. In line with the literature in the field, there was significant heterogeneity in scales employed to report the symptom changes. The description of the sample composition of the included studies also appears inconsistent, with the vast majority providing the gender composition, age range, or the average age of the recruited sample. A discrete heterogeneity also emerged regarding the employed inclusion or exclusion criteria, even considering the heterogeneity of the analyzed diagnostic categories. Numerous different alleles and genotypes have been assessed, but no specific efficacy pattern emerged for a particular marker across the various studies. A relatively limited number of studies [32,34,36,37,39,[41][42][43][44] reported on the results of combinatorial PGx testing, introducing a further layer of complexity in the interpretation of the results for the included studies.

Schizophrenia
Seventeen papers reported on studies comprising individuals affected by SCZ, with only one randomized controlled trial (RCT) [30]. Among them, ten papers reported on the possible association between CYP2D6 and treatment outcomes as described by the authors [30,[45][46][47][48][49][50][51][52], four papers described the association between ABCB1 genotypes and treatment outcomes with three out of four reporting a positive association [47,[52][53][54]. Overall, a significant heterogeneity of assessed outcomes is apparent. Two papers used retention in the treatment of antipsychotics (AP) as the primary outcome [30,55]. As for symptom severity assessment, seven papers used the Positive and Negative Symptoms Scale (PANSS) score as a primary outcome measure, either focusing on total percent change or changes in some of its subscales [45,49,50,52,[55][56][57], whilst seven papers used the Brief Psychiatric Rating Scale (BPRS) percent change [49,51,53,54,[58][59][60]. Seven out of a total of seventeen papers reporting on SCZ described a positive association between PGx markers of efficacy with treatment outcomes [45,48,49,[52][53][54]58]. One study [52] assessed the association of pharmacodynamic together with pharmacokinetic markers of efficacy. An additional paper [60] focused on the association of PGx tests with the change in BPRS-defined cognitive symptoms of SCZ. These results are summarized in Table 1.

Bipolar Disorder
Three papers reported on PGx's association with clinical outcomes in individuals affected by BD. One paper [71] described the association between CYP2D6 and symptom improvement as defined according to the Clinical Global Impression Efficacy Index (CGI-E). An additional paper [32] reported the association of CGI changes with PGx testing of a mixed population comprising BD, PTSD, and MDD. One paper [44] probed the potential cost savings associated with PGx-guided pharmacological therapy changes, focusing on emergency service access. These results are listed in Table 3.

Discussion
A growing amount of evidence points to the potential that PGx holds for treatment personalization in medicine [72][73][74], with notable examples of its applications in cardiology [75], oncology [76], pediatrics [77], and primary care [74], among others. With the right type of information support, PGx may further enhance the shared decision-making between service users and healthcare providers [78]. Great efforts have been invested in testing PGx's efficacy in the pharmacological treatment selection for SMI, and our results seem to confirm our impression regarding its potential value. Meta-analyses of RCTs assessing the effectiveness of gene-guided treatment (GGT) versus treatment as usual (TAU) for MDD point to a modest but statistically significant benefit in terms of a higher remission rate for GGT as compared with TAU [79,80]. However, the clinical adoption of PGx testing in psychiatry appears somewhat delayed [11,23]. Over the years, several reasons have been proposed to explain this phenomenon. Among them, there are a relative lack of RCTs exploring PGx efficacy [11], a lack of knowledge on how to interpret its results by a sizeable portion of healthcare providers [11], inconsistencies in the guidance provided by different clinical practice guidelines [23], and an apparent lack of confidence in the overall value of PGx testing in clinical practice [11,81]. The results of our review seem to point to a significant heterogeneity in assessed outcomes and in the testing panels. Only three papers included in the present project reported on PGx testing in BD, with only one RCT [32]. Considering the current relatively limited number of papers dedicated to the topic, the evidence regarding PGx testing for treatment selection in BD appears particularly scarce. In our data synthesis, less than half of the total studies dedicated to SCZ reported a positive association between PGx and treatment outcomes, and among them three focused on ABCB1 polymorphisms and three additional papers reported on CYP2D6 polymorphisms. The only RCT included in this project and dedicated to assessing PGx testing in SCZ was negative [30]. At this stage, the evidence supporting the use of PGx testing alone to predict treatment outcomes in SCZ does not appear particularly poignant. Blood drug monitoring may represent an additional resource in guiding pharmacological treatment dosing, with clinical practice guidelines specifically dedicated to optimizing their use [82]. Arguably, PGx testing may be synergistically integrated with psychotropic blood monitoring to fully exploit these two different sources of information in optimizing the therapeutic and safety profile for each medication trial. Our study selection did not include any study employing combinatorial pharmacogenomic testing for predicting pharmacological treatment outcomes in SCZ. Fourteen of the twenty-three included studies focusing on MDD described a positive association between PGx testing and pharmacological treatment outcomes [31,33,34,37,39,41,42,61,62,[65][66][67]69,70]. Five of the ten RCTs dedicated to MDD described a positive association for PGx testing and treatment outcomes [31,34,36,37,39], but considering the significant heterogeneity in the testing panels involved, no firm conclusion can be reasonably drawn from our results. Furthermore, a sizeable portion of the available evidence for PGx efficacy presents some financing biases, introducing additional complexity in the overall interpretation of the data [11]. Even pondering the results of the available meta-analyses may be a daunting task, as the proprietary nature of the tested algorithms employed in the involved studies hinders an accurate assessment of the relative impact of each approach [79,83]. Assessing the cost-effectiveness of PGx testing also needs careful consideration and individualized analyses. Commercial PGx costs vary significantly, and there might be differing reimbursement schemes depending on the geographic location with different corresponding healthcare systems and differing frequencies of actionable genotypes in the local population [11]. All these factors lead to the necessity of assessing cost-effectiveness profiles in the specific context where PGx testing should be employed [84,85]. The use of ethnicity as a guiding variable for treatment selection has been subjected to intensified scrutiny during recent decades. However, several clinical practice guidelines use the supposed ethnicity of origin as a possible element on which to base the decision on whether to perform PGx testing or not [86]. Ethnicity-based guidance for screening HLA-B*1502 among individuals of Asian ancestry prior to the use of carbamazepine, as an example, appears misguided and a potential source of confusion as HLA-B*1502 is nearly absent in South Korea and Japan [86]. Indeed, ethnicity represents a poor surrogate for the underlying biology. Therefore, such guidance should be abandoned in favor of more evidence-based, practical screening guidance [86]. Notwithstanding the previously mentioned limitations, a progressive cost reduction and a growing number of tested alleles may expand the number of individuals who may benefit from actionable treatment guidance. These factors, taken together, may increase PGx adoption in clinical practice [23]. Future efforts need to be devoted to improving the standardization for the tested algorithms and clinical practice guidelines, boosting educational programs on how to capitalize on PGx technologies in clinical care and assessing next-generation sequencing in PGx tests to address some of the lasting concerns surrounding PGx use [11,81,87,88].

Limitations
The present paper focused on the association of pharmacokinetic markers, as the available evidence appears to be more solid as compared with pharmacodynamic markers. However, numerous papers have been published on the latter markers, and it would be worthwhile exploring the subject in future review projects. Indeed, the number of published studies on the field is far too great to be covered in a single paper. We did not include papers probing the eventual association between PGx testing and the safety or tolerability of pharmacological treatments. This might have led to the exclusion of a substantial part of the literature and of evidence supporting PGx testing in clinical practice. The search was limited to three databases and to articles written in English, wich could have also impacted on the extensiveness of our analyses. Finally, the lack of consistency in SMI's clinical definition might have hindered our capacity of fully grasping the significance of PGx testing for predicting pharmacological treatment response in psychiatry.

Conclusions
A growing amount of evidence points to the potential that PGx testing holds for improving pharmacological treatment selection in psychiatry. PGx should be seen as an essential tool of an integrated approach which should take advantage of robust and standardized algorithms to help (but not solve) the decision-making process in terms of pharmacological interventions. Another neglected approach is represented by therapeutic drug monitoring, largely underutilized in SMI, but that could further boost the utility of PGx testing if adequately integrated with it. Future efforts will have to address lasting concerns surrounding the lack of standardization of the field and its practical implementation.
Author Contributions: All authors contributed to the development of this paper. P.P., L.B., M.C., M.M. and M.P. performed the first title and abstract screening and a blinded extraction for the study contents. U.I., A.S., C.P., P.P., M.P. and L.B. contributed to elaborating the first draft of the manuscript. M.M., F.P. and B.C. oversaw the whole process and the drafting of the present manuscript. All authors have read and agreed to the published version of the manuscript.