FormalPara Key Points

Natalizumab appears to be correlated to a risk of 0.33 cases/100 persons (95% CI 0.29–0.37) for a median observation period of 44.1 months (IQR 28.4–60) and a mean number of doses of 36.3 (SD ± 20.7). Extended interval dosing carries a lower risk of PML, that is, 0.08 cases/100 persons (95% CI 0.0–0.15) for EID versus 0.3 cases/100 persons (95% CI 0.25–0.34) for SID.

We found a risk of PML after rituximab administration of 0.01 cases/100 persons for a median observation period of 23.5 months (IQR 22.1–42.1), noting the included population is only a part of patients undergoing treatment with this drug.

In our analysis, we found no includible study reporting cases of PML during the course of treatmentwith ocrelizumab, vedolizumab, abrilumab, ontamalimab, teriflunomide, daclizumab, inebilizumab, basiliximab, tacrolimus, belimumab, infliximab, firategrast, disulone, azathioprine or danazole.

1 Introduction

Human polyomavirus 2 (HPyV-2), previously known as JC Polyomavirus (JCPyV or JCV), is a member of the Polyomaviridae family, genus Polyomavirus, isolated for the first time in 1971 in a patient affected by Hodgkin’s lymphoma who died of progressive multifocal leukoencephalopathy (PML), a potentially fatal disease of the central nervous system (CNS) [1, 2]. The viral genome is composed of a 5.13Kb supercoiled circular enclosed double-stranded DNA, containing three regions known as the early coding region, the late coding region and the non-coding control region (NCCR). The early region encodes the alternatively spliced transforming proteins, large tumor antigen (T-Ag) and small tumor antigen (t-Ag), both involved in viral replication. The late region encodes for capsid proteins VP1, VP2 and VP3 and a small regulatory protein, known as agnoprotein, apparently involved in the trafficking of capsid proteins to the nucleus [3, 4].

Early and late coding regions are separated by the NCCR, a genetic sequence essential for viral replication. This sequence can be found in two different forms: a non-rearranged, non-pathogenic form (i.e. archetype form) most frequently found in the urine of healthy subjects, and a rearranged more virulent form (i.e. prototype form) typically detected in the cerebrospinal fluid, brain and blood of PML patients.

The neurotropic pathogenic form is characterized by duplications of the promoter/enhancer elements and deletions of the suppressor elements, these being rearrangements associated with PML development [5, 6].

The virus affects only the human species and has a strong tropism for glial cells, kidney epithelial cells and B lymphocytes. Primary infection takes place by unclear mechanisms; nevertheless, 50–90% of the adult population (range depending on the age and country referred to) presents anti-JCPyV antibodies, suggesting a probable tendency of the virus to cause asymptomatic infections during childhood [5].

JCPyV enters target cells via the serotoninergic 5HT-2a receptor and by binding an N-linked glycoprotein with a terminal α(2,6)-linked sialic acid, both present in a wide variety of human cells. Once JCPyV has entered the cells, a latent infection is established in both kidneys and the lymphoid system. Upon immunosuppression, the virus undergoes a replicative cycle, gaining access to the blood and overcoming the blood–brain barrier [7].

Neurotropic mutants preferably target the oligodendrocytes, where replication provokes an accumulation of viral particles, nuclear enlargement, neuronal apoptosis, and in turn, multifocal demyelination [5].

Although no consensus exists on the various clinical subtypes of PML, Cortese and colleagues referred to specific phenotypes of the disease as ‘classical PML’, ‘inflammatory PML’ or ‘PML IRIS’ [2]. The classical form can involve supratentorial and infratentorial structures, with neurological symptoms depending on the location involved, most commonly represented by cognitive and behavioral abnormalities, sensory and motor deficits, seizures, ataxia, and aphasia. Inflammation-related symptoms and signs could be absent. Conversely, the inflammatory form is usually an expression of an immune reconstitution inflammatory syndrome (IRIS). In some cases, the uncontrolled associated inflammatory response can be fatal during the acute phase. Other pathological manifestations of JCPyV include granule cell neuronopathy, fulminant encephalopathy and JCPyV meningitis [2].

After an initial rise in PML cases described in patients affected by hematological or solid tumors, a second rise was observed during the AIDS pandemic in the 1980s, when people living with human immunodeficiency virus (HIV) represented the vast majority (50–80%) of cases. During the next decades, thanks to the introduction and diffusion of highly active antiretroviral therapy, this patient group progressively decreased in favor of patients affected by autoimmune or neoplastic diseases, or in those undergoing treatment with targeted therapy/immunosuppressants [1, 7]. In the literature, drug-induced PML is most notably correlated with use of natalizumab, a monoclonal antibody acting on α-4 integrin, but a growing amount of data is now available for other immunosuppressants and biologic drugs. To date, multiple sclerosis patients undergoing immunosuppressant treatment represent a large proportion of drug-induced PML cases, but several other drugs have been investigated as possibly correlated to PML and/or mention PML as a risk in their prescription information. Examples include alemtuzumab, brentuximab vedotin, dimethyl fumarate, efalizumab, fingolimod, ibrutinib, obinutuzumab, ocrelizumab, ofatumumab, glatiramer acetate, dalfampridine and rituximab [8, 9].

In this systematic review and meta-analysis, the authors aimed to summarize and assess which drugs are most frequently associated with PML development [8], and report the incidence of drug-induced PML through a meta-analytic approach.

2 Methods

This systematic review and meta-analysis is reported according to the Preferred Reporting Items for Systematic Reviews and meta-Analyses (PRISMA) statement [10]. Details of the protocol for this systematic review were registered on PROSPERO (CRD42022332587).

2.1 Research Question

This systematic review was aimed at answering the question: what is the incidence of drug-induced PML in patients under the age of 80 years, affected by pathologies requiring immunosuppressant treatment, but not affected by HIV, primary immunodeficiencies or malignancies? The question was structured according to the PICOS statement as follows:

Population: people of both sex under the age of 80 years, affected by any pathology but HIV, primary immunodeficiencies or malignancies.

Intervention: exposure to immunosuppressants as indicated for the treatment of their baseline pathologies, regardless the specifics of the latter. Only exposure to drugs known to carry some degree of risk for PML development was considered.

Comparison: the risk of PML was assessed for each drug independently and compared with that of the untreated affected population.

Outcome: estimating the risk of PML development after treatment with each included drug.

Study design: only observational studies and randomized controlled trials (RCTs) were considered eligible for selection.

2.2 Search Strategy and Study Selection

Eligible studies were identified via searching on the following electronic databases: MEDLINE, EMBASE, ClinicalTrials.gov, Web of Science and the Canadian Agency for Drugs and Technologies in Health Database (CADTH).

An initial search was performed on all existing literature up to May 10, 2022, without restrictions in terms of language or publication period. Fully published and peer-reviewed studies were identified by MEDLINE, EMBASE, ClinicalTrials.gov and Web of Science database search, while unpublished studies were sought via the CADTH and selected according to the inclusion criteria. The specific search strategy as elaborated for each mentioned database is available as electronic supplementary material (ESM; S1). After database searching, all results were merged on the online tool Rayyan in order to be deduplicated and selected according to their relevance and the inclusion/exclusion criteria [11]. LVR, RI, DK, NB, IF, FDM, VB and BM all participated to the selection process, which was held, for each record, in blind by at least three authors. Discrepancies in selection were resolved by discussion or by VM, LS or the project coordinator, MI.

As for relevance and eligibility, a first round of elimination was performed in blind by reading the study abstract and title only. Then, a second round of selection was performed by reading the study full text. All phases were supervised by VM and LS.

A detailed flowchart of the selection process is shown in Fig. 1.

Fig. 1
figure 1

PRISMA flowchart—Selection process of included studies

Full size image

2.3 Inclusion and Exclusion Criteria

The inclusion process was limited to RCTs, cohort studies and cross-sectional studies describing or reporting the use of drugs correlated to the development of PML and controlling for PML events over time after administration. Case reports, case series, reviews, or any other study design were considered ineligible. Included populations were any subject up to 80 years of age, who received treatment for at least 3 months with one or more drugs correlated to the development of PML, regardless the pathology requiring such treatment. Subjects older than 80 years of age were excluded, as age itself could represent a factor altering the immune system predisposing to the development of PML [12, 13]. In order to focus the attention on drug-induced cases of PML, people living with HIV, hematological/oncological patients or patients affected by primary immunodeficiencies were considered not eligible for inclusion. Furthermore, we excluded studies reporting administration of mixed or unclear drug regimens, in order to limit possible misinterpretations of the results.

2.4 Data Extraction

Abstracted information was extracted by LVR, RI, DK, NB, IF, FDM, VB and BM, and was entered into a computerized master log. Extracted information included journal, year of publication, first author name, title of the publication, country, study design, study period in months, sample size, sample demographic characteristics, baseline pathology, trial registration number (if applicable), drug name and posology, number of patients developing PML at any time during treatment (number of events), timing of PML development (timing of events), median treatment duration in months, number of drug infusions (median of the cohort), standard/extended interval dosing (for natalizumab only), data about placebo group if present, follow-up period, patient-years and percentage of the population positive to antibodies against JCPyV. Additional data were also gathered for PML cases, such as previous immune-suppressant treatment, number of drug infusions (median), timing of development of PML, treatment strategy for PML cases, information about immune reconstitution syndrome if reported, outcome of PML cases and PML cases in patients undergoing modifications of their treatment regimen. Study investigators were contacted for unreported data or additional details, when needed. Extracted data was checked and elaborated for the meta-analysis by DZ.

2.5 Meta-Analysis

By means of a meta-analytic approach, authors tried to provide a rapid view of the risk of drug-induced PML in patients whose immune system is not additionally depressed by means of neoplasms, HIV or concomitant medications. This was done by analyzing the risk of PML development as reported in the included studies, taking into account drug regimens, exposed population, drug exposure time and relative infusion protocols (e.g. standard interval dosing vs extended interval dosing for natalizumab; different rituximab infusion protocols for its various indications). We excluded from our analysis those studies with promiscuous or unclear drug regimens, such as drug regimens involving two or more drugs both known to carry a risk of PML. This was done to reduce the risk of misinterpreting the results. When median (interquartile range, IQR) of treatment duration was given, we calculated mean and standard deviation (SD) in order to be able to include the studies in the same meta-analysis [14, 15]. Also, when not reported, in order to compare extended interval dosing and standard interval dosing we calculated the mean number of doses based on the treatment duration and dose interval.

The incidence of drug-induced PML was calculated for each drug potentially correlated to PML based on the number of PML cases and total number of patients observed per 100 persons and the observation time as reported in each study. Subsequently, random-effect meta-analyses were conducted for each drug reporting pooled incidence with 95% confidence intervals (CI) and the mean (SD) or median (IQR) of the observation time of all the studies included in the meta-analyses. Forest plots were used for the graphical representation of each meta-analysis.

Heterogeneity was measured by I2 statistic, considering it important when > 50% [16].

Subgroup analysis was also performed for different natalizumab infusion protocols (i.e. standard interval dosing and extended interval dosing) in order to highlight the possible differences in terms of risk of PML development. Publication bias was examined graphically through funnel plots and Egger’s test, where p < 0.05 indicates significant publication bias [17].

The meta-analysis was performed on Stata v.15.0 software (Stata Corp, College Station, TX, USA).

2.6 Quality Appraisal of Included Studies

Each included study was assessed by means of the Cochrane risk of bias tool 2 (RoB2) for clinical trials, or by the Newcastle-Ottawa scale (NOS) for observational studies [18, 19]. Observational studies were divided into very high, high, and low risk of bias, with high risk of bias being assigned to studies scoring between 4 and 6 points on the NOS [19].

The assessment of the risk of bias was performed in blind by IF, VB, BM, DK, FDM and RAC. Disagreements between reviewers was resolved by discussion or by contacting VM, LS or the lead reviewer, MI.

3 Results

3.1 Bibliographic Search

An initial database search on the above-mentioned databases collectively identified 18,573 studies. After merging all records on the computerized tool Rayyan, 7753 duplicates were identified, and, after manual confirmation, removed. After screening the remaining records by title and abstract, 320 were eligible for further assessment by analyzing the full article text. After application of the inclusion and exclusion criteria, 103 studies were selected for the systematic review (Fig. 1).

3.2 Study Characteristics and Included Population

Included studies were published between 2005 and 2022. This review included 76 observational studies [9, 18, 20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94], 26 RCTs [95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120], and data from a pharmacovigilance database [121].

Thirty-two were multicenter studies (31.06%); 15 were based in Italy (14.56%); 11 in the USA (10.67%); 8 in France (7.76%); Germany and Spain contributed with 6 studies each (5.82%); Netherlands contributed with 5 studies (4.85%); Japan with 4 studies (3.88%); and Sweden, Portugal, Switzerland and the UK with 2 studies each (1.94%). Austria, Brazil, Greece, Hungary, Israel, Kuwait, Lebanon and Norway contributed with one study each (0.97%).

Sample sizes of included studies ranged from 9 to 100,921 patients for a total of 228,817 patients.

The mean included population age was 45.51 years (± 12.54). When data about gender was available, male gender represented approximately 27.54% of the included population.

Data about the pathologies discussed in the included studies is depicted in order of frequency in Table 1.

Table 1 Distribution of pathologies described in included studies. Sum is above 103 as the same study may refer to more than one pathology
Full size table

Populations in included studies were most frequently undergoing treatment with natalizumab (n = 65), rituximab (n = 21), fingolimod (n = 10), interferon (n = 14), ocrelizumab (n = 4), vedolizumab (n = 4) and dimethyl fumarate (n = 4). Complete information about all treatment regimens is available in Tables 2 and 3. Given the great number of included studies, the heterogeneity of pathologies and drug regimens, accordingly to our primary objective, and not least, for the sake of maintaining clarity in our manuscript, we decided to focus our discussion on those studies describing at least one case of PML; however, full information about all studies is available in Tables 2 and 3.

Table 2 Overview of included observational studies with information about study design, baseline pathology, investigated drug(s), exposed population, PML cases and extract of Newcastle-Ottawa Scale total score
Full size table
Table 3 Overview of included RCT studies, with information about study design, baseline pathology, investigated drug(s), exposed population, PML cases, extract of RoB2 score
Full size table

3.3 Quality Assessment

All included observational studies were evaluated using the NOS and divided into classes of risk as previously described in the methods section. Two studies were rated as having a very high risk of bias (0–3 NOS points), 59 were rated as having a high risk of bias (4–6) and the remaining studies were rated as having a low risk of bias (7–9). As for the 26 included randomized studies evaluated using the RoB2 tool, 10 were evaluated to have a high risk of bias, 11 showed some concern and five were rated to have a low risk of bias. A detailed chart reporting each subdomain for both the RoB2 tool and the NOS is available in the ESM (S2, S3), while a synthetic view of these data is also available above in Tables 2 and 3.

3.4 Meta-Analysis of Drug-Induced PML Cases

Overall, there were 1017 PML events, reported in 42.71% of the included studies (n = 44). Among these, 70.45% included natalizumab in their treatment regimen (n = 31).

3.5 Natalizumab

A meta-analysis of 41 studies which reported at minimum the number of PML events, total number of patients and observation time showed an incidence of PML cases among patients diagnosed with multiple sclerosis who were undergoing natalizumab treatment according to standard interval dosing of 0.4 cases/100 persons (95% CI 0.36–0.44, I2 = 94%; p < 0.001), for a median observation period of 36 months (IQR 24–60 months) (mean 45.7, SD 28) (Fig. 2). The funnel plot assessing publication bias is presented in Fig. 3. Egger’s test showed a significant small-study effect (p < 0.001).

Fig. 2
figure 2

Meta-analysis for all included studies concerning natalizumab-induced risk of progressive multifocal leukoencephalopathy (PML). The horizontal x-axis is the incidence of PML among referenced studies. Horizontal lines represent confidence interval (CI), black points represent the risk of PML development as calculated for each record. The grey diamond represents the weight of the included study in terms of population. The blue diamond represents the meta-analysis for overall risk of PML development after drug exposure

Full size image
Fig. 3
figure 3

Funnel plot with pseudo 95% confidence limits, assessing publication bias for studies depicted in Fig. 2. The horizontal x-axis is the incidence of progressive multifocal leukoencephalopathy (PML) among referenced studies. s.e. standard error

Full size image

3.5.1 Natalizumab—Mean/Median Time of Treatment and/or Number of Doses Only

When including in the meta-analysis only studies that reported the mean or median time of treatment (n = 25), and/or number of natalizumab doses, the incidence of PML cases was 0.33 cases/100 persons (95% CI 0.29–0.37, I2 = 50.04%; p = 0.003) (Fig. 4) for a median observation period of 44.1 months (IQR 28.4–60 months) (mean 50.3, SD 26.4) and a mean number of doses of 36.3 (SD 20.7). The funnel plot assessing publication bias is presented in Fig. 5. Egger’s test showed a significant small-study effect (p < 0.001).

Fig. 4
figure 4

Meta-analysis for studies concerning natalizumab-induced risk of progressive multifocal leukoencephalopathy (PML), limiting results to those studies reporting treatment duration and/or number of administered doses for the included population. The horizontal x-axis is the incidence of PML among referenced studies. Horizontal lines represent confidence interval (CI), black points represent the risk of PML development as calculated for each record. The grey diamond represents the weight of the included study in terms of population. The blue diamond represents the meta-analysis for overall risk of PML development after drug exposure

Full size image
Fig. 5
figure 5

Funnel plot with pseudo 95% confidence limits assessing publication bias for studies depicted in Fig. 4. The horizontal x-axis is the incidence of progressive multifocal leukoencephalopathy (PML) among referenced studies

Full size image

3.5.2 Natalizumab—Extended Interval Dosing and Standard Interval Dosing

Synthesizing studies that compared extended interval dosing with standard interval dosing, the incidence of PML cases was 0.08 cases/100 persons (95% CI 0.0–0.15, I2 = 0%; p = 0.6) (Fig. 6) for a mean observation period of 43.9 months (SD 21.2) and a mean number of doses of 31.7 (SD 15.8) for extended interval dosing, compared with 0.3 cases/100 persons (95% CI 0.25–0.34, I2 = 22.9%; p = 0.26) (Fig. 7) for a mean observation period of 31.4 months (SD 11.7) and a mean number of doses of 30.0 (9.8).

Fig. 6
figure 6

Meta-analysis for studies concerning natalizumab-induced risk of progressive multifocal leukoencephalopathy (PML) when infused as extended interval dosing. The horizontal x-axis is the incidence of PML among referenced studies. Horizontal lines represent confidence intervals (CI), black points represent the risk of PML development as calculated for each record. The grey diamond represents the weight of the included study in terms of population. The blue diamond represents the meta-analysis for overall risk of PML development after drug exposure

Full size image
Fig. 7
figure 7

Meta-analysis for studies concerning natalizumab-induced risk of progressive multifocal leukoencephalopathy (PML) when infused as standard interval dosing. The horizontal x-axis is the incidence of PML among referenced studies. Horizontal lines represent confidence intervals (CI), black points represent the risk of PML development as calculated for each record. The grey diamond represents the weight of the included study in terms of population. The blue diamond represents the meta-analysis for overall risk of PML development after drug exposure

Full size image

3.6 Fingolimod

A meta-analysis of eight studies showed an incidence of PML cases among patients diagnosed with MS who were undergoing fingolimod treatment of 0.01 cases/100 persons (95% CI 0.00–0.01, I2 = 91.9%; p < 0.001), for a median observation period of 47.6 months (IQR 31.62–53.85 months) (mean 44.5, SD 17.6) (Fig. 8).

Fig. 8
figure 8

Meta-analysis for all included studies concerning fingolimod-induced risk of progressive multifocal leukoencephalopathy (PML). The horizontal x-axis is the incidence of PML among referenced studies. Horizontal lines represent confidence intervals (CI), black points represent the risk of PML development as calculated for each record. The grey diamond represents the weight of the included study in terms of population. The blue diamond represents the meta-analysis for overall risk of PML development after drug exposure

Full size image

When including in the meta-analysis only studies that reported the mean or median time of observation and treatment, the incidence of PML cases was 0.01 cases/100 persons (95% CI 0.00–0.01, I2 = 0.0%; p = 0.820) (Fig. 9) for a mean observation time of 40.9 months (SD 17.4) and a mean treatment duration of 39.2 months (SD 17.9).

Fig. 9
figure 9

Meta-analysis for all included studies concerning fingolimod-induced risk of progressive multifocal leukoencephalopathy (PML), limiting results to those reporting observation time and treatment duration. The horizontal x-axis is the incidence of PML among referenced studies. Horizontal lines represent confidence intervals (CI), black points represent the risk of PML development as calculated for each record. The grey diamond represents the weight of the included study in terms of population. The blue diamond represents the meta-analysis for overall risk of PML development after drug exposure

Full size image

3.7 Rituximab

The meta-analysis of 10 studies showed an incidence of PML cases among patients diagnosed with MS who were undergoing rituximab treatment of 0.01 cases/100 persons (95% CI −0.08 to 0.09, I2 = 20.4%; p = 0.255), for a median observation period of 23.5 months (IQR 22.1–42.1 months) (mean 28.1, SD 12.9) (Fig. 10).

Fig. 10
figure 10

Meta-analysis for all included studies concerning rituximab-induced risk of progressive multifocal leukoencephalopathy (PML). The horizontal x-axis is the incidence of PML among referenced studies. Horizontal lines represent confidence intervals (CI), black points represent the risk of PML development as calculated for each record. The grey diamond represents the weight of the included study in terms of population. The blue diamond represents the meta-analysis for overall risk of PML development after drug exposure.

Full size image

3.8 Dimethyl Fumarate

A meta-analysis of three studies showed an incidence of PML cases among patients diagnosed with MS who were undergoing dimethyl fumarate treatment of 0.17 cases/100 persons (95% CI 0.12–0.22, I2 = 64.4%; p = 0.06), for a median observation period of 36.1 months (IQR 24–77.8 months) (mean 45.9, SD 28.2) (Fig. 11).

Fig. 11
figure 11

Meta-analysis for all included studies concerning dimethyl fumarate-induced risk of progressive multifocal leukoencephalopathy (PML). The horizontal x-axis is the incidence of PML among referenced studies. Horizontal lines represent confidence intervals (CI), black points represent the risk of PML development as calculated for each record. The grey diamond represents the weight of the included study in terms of population. The blue diamond represents the meta-analysis for overall risk of PML development after drug exposure.

Full size image

4 Discussion

In this study, we reviewed and meta-analyzed the risk of drug-induced PML in subjects affected by pathologies requiring immunosuppressive treatments, but not additionally immunocompromised by means of neoplasms, HIV or concomitant medications. This study intended to take into account globally drug-induced PML as reported by available literature, including a wide range of baseline pathologies and relative treatments, such as those affecting the gastrointestinal system, the central nervous system, rheumatologic disorders and in general diseases caused by dysregulation of the immune system.

Studies reporting the use of unclear or combined therapy were excluded from the meta-analysis such as in the case of interferon, for which we found no eligible case of PML reported, unless co-administered or administered shortly after a natalizumab-based regimen. Furthermore, no included study reported any case of PML during follow-up or treatment with ocrelizumab, vedolizumab, abrilumab, ontamalimab, teriflunomide, daclizumab, inebilizumab, basiliximab, tacrolimus, belimumab, infliximab, firategrast, disulone, azathioprine or danazole. However, literature reports cases of PML during treatment with one or more of the above-mentioned drugs, such as in the case of ocrelizumab, tacrolimus or inebilizumab [122, 123], the latter being listed as cause of a possible PML case in a deceased patient during an inebilizumab clinical trial [124]. However, these and similar reports were not included in the present study as they did not meet our inclusion criteria.

As for dalfampridine, all 50 cases of PML found were described by a single study [9]. Nevertheless, as the authors clarify in their study, 46 of 50 PML patients were previously treated with natalizumab, making it possible, if not likely, that there was a carryover effect after natalizumab administration. On a similar note, the same study also reports 24 cases of PML during or after glatiramer acetate treatment. We decided to exclude data about these two drugs from our meta-analysis, considering also their mechanism of action [125, 126]. Nevertheless, we included all data from the mentioned study in the systematic review (Tables 2, 3).

When considering the risk of PML in the drug-exposed population, dalfampridine, glatiramer acetate, dimethyl fumarate and fingolimod show a relatively safe profile, with only a few studies reporting events of PML in the treated population, mostly due to carry-over risk from previous treatment. Regardless of the specifics of each single included report, the meta-analysis reduces the risk of drawing misguided conclusions, even when data were lacking, such as in the case of records not reporting the entire treatment history for all included patients.

Besides that already stated for dimethyl fumarate, it is worth mentioning that the reported results may be largely influenced by the effect of single studies, and even though this drug appears relatively safe in terms of PML events, we could not obtain the necessary information about the specific known risk factors for the development of PML during dimethyl fumarate treatment, such as absolute lymphocyte count for all cases [127]. The sub-analysis for this kind of data, although fundamental, could be applicable mutatis mutandis to all discussed drugs. Nevertheless, this appeared beyond the purpose and design of our study.

As for rituximab, our results apply in the limited setting of our included population. This may be a limitation, since our results may not be generalizable to the whole population undergoing treatment with rituximab, but it could also provide an interesting point of view about this drug, as our data focus solely on the drug influence on PML risk. This may be useful in a comparison of this risk with that of the hematological population undergoing rituximab treatment, where both the drug and the pathology itself may act as a strong immunosuppressive stimulus, possibly in a synergistic manner. Once more, this aspect was beyond the purpose of our study.

As previously mentioned, we could not include single studies in the meta-analysis. This is the case for the nine and four cases of PML described after treatment with mycophenolate mofetil and belatacept, respectively. Nevertheless, data about these cases is available as part of our systematic review (Tables 2, 3).

Our meta-analysis showed treatment with natalizumab carries a risk for the development of PML of 0.33 cases/100 persons (95% CI 0.29–0.37), for a median observation period of 36 months (IQR 26.1–60 months) (mean 47.7, SD 25.7) and a mean number of doses of 33.5 (SD 14). Nevertheless, the FDA Adverse Event Reporting System (FAERS) counts a total of 1914, which is more than the cases reported in our systematic review and meta-analysis [128]. Reasons for this discrepancy include ineligible study design, lack of clinical data (e.g. treatment duration, mean number of doses administered), lack of published reports, ineligible population, presence of confounders such as carry-over therapy, advanced age and other immunosuppressive treatment. According to the extremely limited data available, it could be safe to assume there is a difference in risk according to the infusion protocol used, as depicted by our meta-analysis comparing data about patients treated with standard interval dosing and extended interval dosing. When risk of PML is calculated among extended or standard infusion dosing patients only, extended interval dosing-related risk appears to be smaller, that is, 0.07 cases/100 persons (95% CI 0.0–0.15) for a mean observation period of 48.2 months (SD 19.6) and a mean number of doses of 35.1 (SD 14.5) compared with 0.3 cases/100 persons (95% CI 0.25–0.34), for a mean observation period of 33.7 months (SD 11.1) and a mean number of doses of 32.0 (SD 9.1). The compared efficacy of these two regimens and clinical response evaluation was beyond the purpose of the present study. Our search strategy was designed to include all solid organ transplant patients. Nevertheless, few records, all concerning kidney transplant patients, fulfilled our inclusion criteria and reported useful data for our analysis. We decided to proceed with our analysis of the eligible records in accordance with our initial inclusion criteria even though we acknowledge the population of solid organ transplant patients might be underrepresented.

4.1 Strengths and Limitations of the Present Study

To our knowledge, this is the first systematic review and meta-analysis aimed at collectively describing all drug-induced PML in individuals exposed to immunosuppressants, but whose immune system is not additionally suppressed by means of neoplasms, concomitant medications or HIV, without restricting the search to specific drugs or to specific pathologies. In order to do so, great effort was made to retrieve all relevant literature from inception, in all languages. In order to include high quality research, we excluded case reports and case series from our analysis. However, as a result, the occurrence of PML during treatment with some of the above-mentioned drugs was not included. At the same time, this could represent both a strength and a limitation of the present study. In this work we considered as immunocompetent any patient without solid organ tumor, hematologic malignancies, HIV, or primary immunodeficiency. Nevertheless, we acknowledge some autoimmune disease, such as sarcoidosis, could itself be associated with a certain degree of autoimmune dysregulation [129].

The present study also presents several limitations that are important to discuss to ensure a better understanding of the results. Firstly, data from included studies were frequently insufficient or unclear. This was particularly limiting in the case of observation time. While some studies reported observation time after drug administration as follow-up time, others ambiguously reported follow-up data as ‘treatment duration’, ‘study period’ or other unprecise indications. Another example of this issue was represented by the lack of information about JCV serostatus/JCV index in included literature. When extracting such data, we found it to be sparse and most often lacking, making it impossible to include and to meta-analyze. Even if the present record mentions no information about this issue, we acknowledge its clinical relevance in PML risk calculation and invite the reader to consider it as a limitation of the review [12, 130]. When data were unclear, we used the safest assumption available. Nevertheless, this may have altered the results to an unpredictable extent. Concerning reported cases of drug-associated PML, discrepancies in numbers between our results and those reported elsewhere could be due to multiple reasons, such as ineligibility of study design, lack of clinical pharmacological data, unavailability of published reports, ineligibility of population due to confounding factors such as age and previous or multiple immunosuppressive treatment. Furthermore, many records had to be excluded due to the lack of an overall drug-exposed population, making an assessment of the risk impossible. These records were therefore excluded from our analysis.

Another element to consider is that malignancies were excluded from the present review and meta-analysis. While this was intended to focus the attention solely on the contribution of immunosuppressants in the development of PML, it is worth considering that results found in the present record may not reflect either the totality, nor the majority of the patients undergoing treatment with drugs such as rituximab, most often used in a very diverse subset of patients from the one included here.

Funnel plot analysis found publication bias, which could be a result of having found, and therefore included, only peer-reviewed studies. Otherwise, this could result from the heterogeneity of included studies. It could be useful to report, though, that a great proportion of data was obtained by registrative studies run by pharmaceutical companies. However, as previously discussed by Page et al., funnel plots should be seen as a generic means of examining small study effect, hence the tendency for the smaller studies in a meta-analysis to show larger treatment effects, and not as a tool to diagnose bias, since several factors may lead to asymmetry in a funnel plot [131].

5 Conclusions

A high risk of drug-related PML in the not otherwise immunosuppressed population is found in the neurological field. This risk is higher during multiple sclerosis treatment, and highest during long-term natalizumab therapy. Although this drug is still routinely prescribed in this field, in other areas, such as inflammatory bowel diseases, it could play a progressively smaller role, and be gradually replaced by other more recently approved agents, such as vedolizumab, for which we found no reported case of PML. In other areas of medicine, where the use of modern targeted therapy already plays an important role, the risk of drug-induced PML is much less permeating.