1. Data source and report identification

This study utilized data from the US Food and Drug Administration Adverse Event Reporting System (FAERS) database [21] (https://www.fda.gov/drugs/drug-approvals-and-databases/fda-adverse-event-reporting-system-faers-database), including all spontaneously reported adverse drug event records worldwide from the first quarter of 2004 to fourth quarter of 2024. The FAERS database contains multiple tables, including demographic (DEMO), drug (DRUG), indication (INDI), adverse reaction (REAC), and outcome (OUTC) information.

Reports of PML were identified using the MedDRA Preferred Term “Progressive multifocal leukoencephalopathy.” Identified reports underwent validation through review of narrative fields and supporting documentation by two independent reviewers, with disagreements resolved by a third reviewer. Reports with incomplete information or where PML could not be confirmed were excluded from the analysis.

2. Drug and indication classification

All drugs were standardized to International Nonproprietary Names (INN), taking into account possible synonyms and spelling variations. Vaccines, diagnostic agents, and combination preparations were excluded. Herbal medicines and records with unclear drug indications were also excluded. Indications were categorized based on the MedDRA classification system into major underlying disease groups: autoimmune diseases (including multiple sclerosis, rheumatoid arthritis, and inflammatory bowel disease), malignancies (hematological and solid tumors), organ transplantation, and other conditions. For reports with multiple indications, the primary underlying disease was determined based on the reported indication for the suspected causative drug, clinical context, and medical literature.

3. Signal detection methods

Four mainstream pharmacovigilance signal detection methods were jointly applied, including two frequentist disproportionality measures (ROR [22,23] and PRR [24,25]) and two Bayesian shrinkage approaches (BCPNN [18,26] and MGPS [19,26]). ROR/PRR are derived from 2 × 2 contingency tables and are intuitive to interpret but can be sensitive to sparse counts, whereas BCPNN/MGPS use Bayesian shrinkage to stabilize estimates for rare drug–event pairs. We therefore required concordant signals across all four methods to prioritize specificity and reduce spurious findings due to random variation, acknowledging a potential loss of sensitivity [27].

• ROR: ROR > 1 and 95% confidence interval (CI) lower bound > 1;
• PRR: PRR > 2and χ² > 4 and N ≥ 3;
• BCPNN: Information Component (IC025) > 0;
• MGPS: EBGM05 ≥ 2.

4. Ethics statement

This study used publicly available, anonymized data from the FAERS database and did not involve direct patient intervention or identifiable personal information. This research qualified for exemption from ethics committee approval, consistent with previous pharmacovigilance studies using FAERS data [28]. Data access and analysis complied with the FDA’s FAERS data use policies and applicable privacy regulations.

1. Data screening and reports summary

From the FAERS database, a total of 18,613,992 unique patient records were extracted from the first quarter of 2004 to the fourth quarter of 2024 after data deduplication and multi-table integration. Further screening resulted in 7,244 drug-PML pairings, involving 298 drugs potentially associated with PML. By applying four signal detection algorithms (ROR, PRR, BCPNN, and MGPS), the top 30 drugs ranked by the number of reports are presented in Table 1, with the complete list provided in S1 Table. The complete data processing and screening flow is presented in Fig 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Top 30 (of 72) drugs showing positive PML signals in all four detection algorithms, ranked by report numbers. (complete list provided in the S1 Table).

https://doi.org/10.1371/journal.pone.0341855.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Drug signal detection process for progressive multifocal leukoencephalopathy in FAERS.

https://doi.org/10.1371/journal.pone.0341855.g001

2. Characteristics of reported PML reports in FAERS

A total of 7,244 PML reports were included in this study. The overall mean age of patients was 53.8 ± 15.6 years, with a mean body weight of 70.4 ± 18.8 kg. The overall male-to-female ratio was approximately 1:1.3, with females slightly predominant in reports. Subgroup analysis revealed that reports associated with HIV/AIDS were predominantly male, whereas females were notably higher in non-HIV/AIDS-associated reports (Fig 2e).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Clinical and demographic characteristics of progressive multifocal leukoencephalopathy reports in FAERS (2004–2024).

a: Annual number of PML reports in HIV versus non-HIV populations. b: Clinical outcomes of PML reports, including proportions of life, death, hospitalization, disability, and missing data. c: Geographic distribution of PML reports. The map was created by the authors using Datawrapper (https://app.datawrapper.de/) with a public-domain base map from Natural Earth. d: Reporter category (e.g., physician, other healthcare professional, consumer). e: Proportion of HIV and non-HIV PML reports by sex. Abbreviations: HIV, human immunodeficiency virus.

https://doi.org/10.1371/journal.pone.0341855.g002

Reporting of PML to FAERS has increased substantially in recent years, while HIV/AIDS-related reports remained stable at 20–30 per year. Non-HIV/AIDS reports now account for several hundred annually over the past decade (Fig 2a). These trends reflect reporting patterns and drug utilization changes rather than disease incidence. More than a third of reported cases documented life-threatening complications or death (Fig 2b). Geographically, Europe and North America contributed the most reports (Fig 2c). Physicians and healthcare professionals were the primary reporters of these reports (Fig 2d). These patterns reflect reporting behaviors and healthcare system characteristics rather than true PML epidemiology.

3. Results of drug signal detection

Our pharmacovigilance analysis identified 72 drugs showing disproportionate PML reporting patterns using four independent signal detection algorithms (Table 1 and S1 Table). Fig 3 presents the forest plot of reporting odds ratios (RORs) with 95% confidence intervals for these agents, arranged in descending order based on the report frequency.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Forest plot of RORs with 95% CI for 72 drugs positively associated with progressive multifocal leukoencephalopathy.

BIKTARVY, Bictegravir & Emtricitabine & Tenofovir Alafenamide; ATG, Antithymocyte Immunoglobulin; STRIBILD, Cobicistat & Elvitegravir & Emtricitabine & Tenofovir; GENVOYA, Cobicistat & Elvitegravir & Emtricitabine & Tenofovir Alafenamide. Abbreviations: ROR, reporting odds ratio; CI, confidence interval.

https://doi.org/10.1371/journal.pone.0341855.g003

Among the signal-positive drugs, natalizumab and rituximab had the highest reporting volumes (1,848 and 1,298 reports, respectively) with corresponding ROR values of 38.08 (95% CI: 36.11–40.15) and 39.44 (95% CI: 37.13–41.9). These were followed by mycophenolic acid (271 reports; ROR: 12.13, 95% CI: 10.74–13.7) and fingolimod (249 reports; ROR: 8.01, 95% CI: 7.06–9.09). The plot demonstrates substantial heterogeneity in reporting associations across drugs, with ROR values ranging from 2.91 (95% CI: 2.04–4.17) for ofatumumab to 40.42 (95% CI: 24.65–66.26) for chlorambucil.

4. Analysis of drug indications and disease classification

4.1. Pharmacological classification of PML-related drugs.

As shown in Fig 4, we classified the 72 PML-associated drugs into five major pharmacological categories. The analysis revealed that immunomodulatory drugs represented the largest proportion, which were further divided into three subcategories: drugs targeting lymphoid B and T cells (3,443 reports, approximately 60% of all reports), broad-spectrum immunosuppressants (551 reports), and small molecule immunomodulators (468 reports). The other major categories included hormone drugs (424 reports), cytotoxic drugs (728 reports), therapeutic drugs (103 reports), and antiviral agents (279 reports). The pie chart clearly illustrates the distribution of these drug categories, with the blue section representing the major category of immunomodulatory drugs, which includes the three subcategories mentioned above. Collectively, these three subcategories account for approximately 77% of all PML-related reports.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Composition and distribution of 72 progressive multifocal leukoencephalopathy -associated drugs by pharmacological category.

The blue section indicates a major category, which includes three subcategories: drugs targeting lymphoid B and T cells, broad-spectrum immunosuppressants, and small molecule immunomodulators.

https://doi.org/10.1371/journal.pone.0341855.g004

5. Onset timing and risk characteristics of major drugs

Among the 72 signal-positive drugs, the top 10 were selected for analysis of time-to-onset (TTO) from drug initiation to PML reporting. Median TTO varied widely across drugs, ranging from 49 days (ocrelizumab) to 1343 days (dimethyl fumarate). Ocrelizumab had the shortest median TTO (49 days), followed by natalizumab (262 days) and rituximab (335 days), while other drugs showed incrementally longer intervals (Fig 5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Time-to-onset analysis of ten high-risk progressive multifocal leukoencephalopathy -associated drugs.

https://doi.org/10.1371/journal.pone.0341855.g005

1. Trends in PML epidemiology and changes in population characteristics

PML was initially recognized as being most common among patients with intrinsic immune deficiencies, such as those with HIV/AIDS. With the widespread adoption of highly active antiretroviral therapy (HAART), the number of HIV-associated PML reports has stabilized at a relatively low level [6,12]. However, over the past two decades, the extensive use of a large number of immunosuppressive drugs—including biologics and immunomodulators—in the fields of cancer, autoimmune diseases, and organ transplantation has led to a rapid increase in drug-related PML [29–32]. The composition of PML reports in FAERS has shifted markedly, with HIV/AIDS-associated cases remaining low while drug-related PML reports have progressively increased to predominate (Fig 2a). This study reflects that drug-associated PML has become a new clinical focus for prevention and control. From the perspective of geographic distribution (Fig 2c), PML reports are predominantly concentrated in Europe and North America, likely reflecting stronger pharmacovigilance systems and widespread use of immunomodulatory agents in these regions [33], though reliance on MRI/cerebrospinal fluid JCV PCR for diagnosis [1] and the US-centric nature of the database may affect case confirmation and reporting completeness. Most reports in the FAERS database come from healthcare professionals (Fig 2d). This enhances the reliability of our data compared to spontaneous reporting systems that rely primarily on patient self-reporting. Regarding sex differences [34], we observed a higher proportion of female patients in the non‑HIV population. This likely reflects underlying disease epidemiology and treatment patterns—particularly higher prevalence of systemic lupus erythematosus and rheumatoid arthritis in women, leading to greater exposure to immunosuppressive/immunomodulatory therapies and more reported PML cases [35]—rather than a true sex‑specific drug effect. Although sex‑related biological factors in JCV infection and PML pathogenesis have been discussed [36], FAERS is subject to reporting bias, missing data, and unmeasured confounding; our findings represent disproportionality signals and do not permit causal inference on sex‑specific susceptibility.

2. Distribution of drug signals and analysis of high-risk indications

In this study, four mainstream signal detection methods (ROR, PRR, BCPNN, MGPS) were jointly used to determine risk drugs, including only those identified as positive by all four, thereby ensuring scientific rigor and reliability in the analysis of drug signals associated with PML. A total of 72 PML risk drugs were ultimately identified, mainly covering immunomodulators, cytotoxic drugs, biologic agents, and some novel targeted drugs (Fig 4). Our analysis revealed substantial heterogeneity in PML reporting odds ratios (RORs) among immunomodulatory agents, consistent with previous pharmacovigilance studies. Oshima et al. (2019) identified the highest reporting associations for natalizumab, while fingolimod showed markedly lower adjusted odds ratios [37]. Similarly, Honma et al. (2024) reported elevated RORs for B-cell–targeted agents such as rituximab, likely influenced in part by enhanced reporting awareness following established warnings [38]. Sharma et al. (2022) emphasized that carry-over effects from prior natalizumab exposure might contribute to PML occurrence during subsequent rituximab therapy [39]. Hence, FAERS-derived ROR signals should be interpreted as indicators of reporting association rather than absolute incidence, and integrated with clinical and epidemiological evidence for comprehensive risk evaluation.

Heatmap (S1 Fig) analysis of high-signal drugs and their indications showed that higher reporting rates of PML were highly concentrated in patients with multiple sclerosis, lymphoma, plasma cell myeloma, rheumatoid arthritis, and other immune-related diseases. S2 Fig further consolidates the top 30 diseases into 17 disease categories, also showing the correlation between immune and tumor-related diseases and PML. These patients are prone to immunosurveillance deficits either due to the diseases themselves or prolonged immunosuppressive therapy.

However, the observed reporting patterns are influenced by multiple biases: neurologists and hematologists have higher PML awareness and may report cases more frequently than rheumatologists or dermatologists; early drug safety warnings (e.g., natalizumab black box warning) may also have increased reporting [37,38]. Therefore, the concentration patterns reflect the combined result of true susceptibility, clinical practice, and reporting behavior, not equivalent to actual incidence risk distribution.

It is worth noting that among certain malignancies such as leukemia, PML may also occur even in the absence of immunosuppressant use because of persistent lymphocytopenia secondary to the disease itself [40,41]. This suggests that, in such cases, PML is often attributable to the synergistic immunosuppressive effects of both the underlying disease and drug therapy. By contrast, for patients with multiple sclerosis or post-organ transplant, PML more commonly results from long-term or intensive immunosuppressive treatment alone.

3. Pathogenic mechanisms and risk features of key drugs

While FAERS data are derived from a passive surveillance system and do not provide incidence estimates, we sought to explain high-signal drug–PML associations from a plausible pathogenic mechanisms perspective. Immunomodulators—such as rituximab and natalizumab—affect the body’s immune status by acting on immune cells such as B/T cells. Previous studies have shown that low CD4+ T lymphocyte counts can be used for PML risk stratification [11,42]. However, natalizumab and rituximab, which ranked in the top 2 by ROR in our study, do not directly act on CD4+ T lymphocytes, and peripheral CD4+ T lymphocyte counts actually increase during natalizumab treatment [43], which seems inconsistent with previous research conclusions. Further review of the literature revealed that CD4+ T lymphocytes decrease during rituximab maintenance therapy, and CD4+ T counts show an upward trend when maintenance is discontinued, suggesting a synergistic effect between B/T cells [44–46]. Natalizumab-associated PML cases are time-dependent, and compared to CD4+ T cell counts, the decrease in CD4+ T lymphocyte bioenergetics (such as intracellular ATP) in natalizumab-associated PML cases may better reflect the body’s immune status [47]. In addition, the decrease in the frequency of specific CD4+ T subsets (central memory CD4+ T cells) is more closely associated with PML occurrence [48]. In summary, CD4+ T lymphocyte assessment should adopt a multidimensional strategy, fully considering CD4 counts, bioenergetic status (intracellular ATP concentration), and functional subsets.

Cytotoxic drugs—including alkylating agents, antimetabolites, cytotoxic chemotherapy drugs, etc.—induce cytotoxicity by interfering with DNA/RNA synthesis and replication, eliminating rapidly dividing cells, and significantly inhibiting the growth and proliferation of immune cells, causing sustained or profound immunosuppression [1,12]. Glucocorticoids suppress immune function via multiple mechanisms, with long-term or high-dose use markedly increasing PML risks [49].

HAART restores immunity and reduces opportunistic infections [50]. During this process, immune reconstitution may trigger immune reconstitution inflammatory syndrome (IRIS) [51–53], which is a paradoxical inflammatory response causing worsening of existing PML or new-onset disease with high morbidity and mortality [54].IRIS occurs during antiretroviral therapy or following immunosuppressant withdrawal [55].

In addition, this study has found associations between chloroquine, hydroxychloroquine, and mirtazapine with PML reports. Currently, there is no standard treatment for PML, although drugs such as mirtazapine, mefloquine, cidofovir, recombinant interleukin-7, interferon-α, and intravenous immunoglobulin (IVIG) are being explored [14,56,57]. Case reports and basic research suggest that JCV infection of oligodendrocytes depends on serotonin 2A receptor activity; thus, 5-HT2A receptor antagonists such as mirtazapine may be effective [58]. In vitro experimental evidence shows that mefloquine can inhibit JCV replication, though clinical efficacy remains limited [59,60]. Consequently, the principal goal in the treatment of PML remains restoring sufficient immune function to control JCV [1].

FAERS data cannot effectively disentangle drug effects from baseline risk (including underlying indication, disease severity, and concomitant immunosuppression), nor can it distinguish between therapeutic drugs and causative drugs; therefore, these signals should be viewed as hypothesis-generating rather than causal evidence, which constitutes an important limitation of this analysis.

4. Median time to onset (TTO) analysis of major drugs

We observed substantial heterogeneity in reported TTO (Fig 5), ranging from 49 days to 1343 days (approximately 1.6 to 44.8 months). Our findings differ from Maas’s study reporting average TTO of 29.5 months (14.2–37.8 months in subgroups) [10].

Our relatively shorter TTO may be attributed to several factors. First, data incompleteness inherent in spontaneous reporting systems may affect the accuracy of temporal analyses, as highlighted by Hauben and Aronson [61], who demonstrated that missing data and reporting biases could significantly impact time-to-event estimates in pharmacovigilance databases.. Second, the heterogeneity in baseline diseases among our cohort likely influences PML onset timing, as Berger et al. [62]. Third, as illustrated in our analysis and consistent with Maas et al.‘s findings [10], different immunosuppressive mechanisms produce varying time-to-onset patterns, with some drugs like natalizumab typically showing longer latency periods compared to other agents. Furthermore, Williamson and Berger [63] have described how different immunomodulatory mechanisms directly influence the timeline of PML development, with lymphocyte trafficking inhibitors showing distinct temporal patterns compared to B-cell depleting agents. Finally, differences in prior drug exposure may significantly impact TTO, as Schwab et al. [64] demonstrated that previous immunosuppressive treatment history can accelerate PML development, potentially explaining some of the shorter TTO reports in our cohort. In summary, TTO explanation needs to be more cautious.