This retrospective, case-control study was conducted at La Paz University Hospital in Madrid, Spain, a tertiary-care teaching facility. Since 2007, all admissions have been monitored by the Proactive Pharmacovigilance Program from Laboratory Signals in Hospital to proactively detect serious ADRs (Ramirez et al., 2010). The study, conducted between 2018 and 2024, investigated thirty-nine suspected cases of drug-induced cytopenia detected by the pharmacovigilance program or through consultations from other specialties within the Pharmacovigilance Unit of the Clinical Pharmacology Department.

The study was approved by La Paz University Hospital Ethics Committee (Code PI-3226; 25 May 2018). Due to the retrospective nature, the study was exempt from informed consent requirements. For all patients initially categorised as having suspected drug-induced cytopenias (neutropenia, agranulocytosis, bone marrow aplasia or haemolytic anaemia), a complete report was submitted to the pharmacovigilance centre in Madrid, Spain (https://www.notificaram.es).

The inclusion criteria for the blood cytopenia cases were: 1) meeting the definition of any type of cytopenia (neutropenia, agranulocytosis, haemolytic anaemia, bicytopenia or pancytopenia), 2) prior drug intake before the blood cytopenia index date, 3) reasonable exclusion of all alternative causes according to the SPS algorithm (Aguirre and García, 2016), and 4) at least one drug had a SPS score ≥+4. Medication errors were excluded from the study.

For agranulocytosis cases, the following criteria standardized by Benichou et al. (Benichou and Solal Celigny, 1991) were also applied: 5) onset of agranulocytosis occurred during treatment or within 7 days of the previous intake of the same drug and no clinical features and more than 1.5 × 10⁹ neutrophils/mm³, 1 month after drug discontinuation; 6) the absence of the exclusion criteria “history of congenital or immune-mediated neutropenia,” “recent infection (especially viral),” “prior chemotherapy or radiotherapy” or “therapy with biologicals, haematological disease” and, 7) recurrence of neutropenia or agranulocytosis after repeated treatment with the drug (positive rechallenge).

Age- and sex-matched patients who completed the drug therapy without experiencing any adverse reactions during the study period served as tolerant controls.

Patient safety was ensured by promptly discontinuing the suspected drug(s) after excluding alternative causes, along with close clinical monitoring within the Pharmacovigilance Unit. When drug re-exposure was clinically indicated, it was performed under strict medical supervision with appropriate laboratory follow-up.

The procedure of the pharmacovigilance program for detecting drug-induced cytopenia cases has been described elsewhere (Ramirez et al., 2010). Briefly, in phase I, on-file laboratory data at admission or during hospitalisation were screened 7 days a week, 24 h a day, for results of abnormally low values of blood cell count. In phase II, the patients were identified to avoid duplicates, and electronic medical records were reviewed. In those cases, where low blood cell count was clearly attributable to other alternative causes, the patients were not further analysed because a drug-induced cytopenia was unlikely. In phase III, a case-by-case evaluation was performed for the remaining cases. When the drug history was unclear, we interviewed the patients or their relatives to obtain more details and conducted additional tests.

For each case, the index day was defined as the earliest occurrence of either the onset of clinical symptoms or the detection of cytopenia in blood tests.

Once ruled out alternative causes for blood cytopenias, the suspicious drugs were withdrawn after discussion with the attending physician and patients were offered to be followed in the Pharmacovigilance Unit.

The case definition of the different types of IDIN relied on the following clinical chemistry criteria: 1) neutropenia: ANC < 1.5 × 10³/µL, Hb ≥ 10.0 g/dL and APC ≥ 0.1 × 10⁶/µL; 2) agranulocytosis was defined as ANC < 0.5 × 10³/µL, Hb ≥ 10.0 g/dL and APC ≥ 0.1 × 10⁶/µL; 3) haemolytic anaemia: Hb < 6.5 g/dL with reticulocyte count > 0.105 × 10⁶/µL; LDH levels > 190 UI/L, UCB > 1.2 mg/dL, WBC count ≥ 3.5 × 10³/µL and APC ≥ 0.05 × 10⁶/µL; 4) bicytopenia: WBC count ≤ 3.5 × 10³/µL and one of the following: Hb < 10.0 g/dL or APC ≤ 0.05 × 10⁶/µL; 5) pancytopenia: WBC count ≤ 3.5 × 10³/µL, Hb ≤ 10.0 g/dL, APC ≤ 0.05 × 10⁶/µL.

The causality assessment was performed using the causality algorithm of the SPS (Aguirre and García, 2016) (Supplementary Figure S1). This algorithm evaluates the causal relationship between a suspected drug and ADR. The SPS algorithm consists of 7 questions or criteria: 1) time between the start of treatment and ADR onset; 2) prior evidence in the literature regarding the relationship between the drug and the ADR; 3) ADR evolution after drug withdrawal; 4) effect of re-exposure; 5) presence of non-drug or other drug related cause; 6) presence of contributing factors favouring causality relationship, and 7) results of tests supporting drug causality.

The total score (ranging from −8 to +11) from the domain-specific assessment classifies the drug into 5 separate categories: definite (≥8), probable (6–7), possible (4–5), conditional (1–3) or unrelated (<0). A SPS score ≥+4 was considered drug-related.

Drugs implicated in the causation of drug-induced blood cytopenias were classified into therapeutic subgroups according to the Anatomical Therapeutic Chemical (ATC) system. Latency periods reported in the literature were recorded and compared with observed latencies.

LTT was performed using different concentrations of the drug(s) involved in the blood cytopenias cases (SPS algorithm score ≥+3) and tolerant controls in order to assess its contribution in enhancing causality assessment, although only drugs with a final score ≥+4 were classified as “suspected culprit drugs” in the analysis. LTT was performed after cytopenia recovery and at least 1 month after steroid therapy was stopped, if applicable.

Lymphocyte proliferation was measured as previously described (Pichler and Tilch, 2004; Vílchez-Sánchez et al., 2020a). Briefly, mononuclear cells were separated over a density gradient (Histopaque-1077, Sigma-Aldrich) from fresh peripheral blood and were plated in flat bottom wells of microtitre plates at 2 × 10⁵ cells/well. Cells were incubated for 6 days with various drug concentrations in triplicate. Drugs were assayed at concentrations of 1, 10 and 100 μg/mL, and occasionally, a lower or higher concentration (0.1, 200 or 500 μg) were used, as previously described (Pichler and Tilch, 2004). We used phytohemagglutinin (5 μg/mL) as a positive control. For the final 18 h of the incubation period, proliferation was determined by adding 1 μCi [3H] thymidine.

All laboratory procedures involving human peripheral blood samples were conducted in accordance with biosafety level 2 (BSL-2) conditions. Institutional protocols were strictly followed for sample handling, storage, and disposal to ensure both staff and environmental safety.

Proliferative responses were expressed as the stimulation index (SI), calculated as the ratio between the mean counts per minute in cultures with the drug and those without drug. A receiver-operating characteristic (ROC) curve analysis was performed using LTT results from 85 drug-tolerant participants to determine the optimal SI cut-off value. An LTT result was considered positive if the SI exceeded the threshold at any drug concentration. Patients were classified as having immune-mediated cytopenia if at least one LTT result was positive.

Continuous variables are presented as mean ± standard deviation (SD) or median with interquartile range (IQR), depending on normality assessed by the Kolmogorov-Smirnov test. A comparison was made between the latency described in the literature and that observed in the study. Categorical variables are expressed in absolute terms and percentages. We employed the chi-squared test to compare the categorical variables and employed Student’s t-test for the continuous variables with a normal distribution. In the event the data did not have a normal distribution, we used the nonparametric Mann-Whitney U-test or Kruskal-Wallis test, as appropriate. Differences were considered significant when the p-value was <0.05. The strength of the association between female sex and the development of cytopenia was quantified using an Odds Ratio (OR) with its 95% confidence interval (CI). A p-value of less than 0.05 was considered statistically significant. Descriptive analysis was performed using R (Integrated Development for R Studio, PBC, Boston, MA; http://www.rstudio.com, accessed on 19 February 2025).

Receiver-operating characteristic (ROC) curve analysis was performed to determinate the optimal cut-off value for the SI in LTT. This analysis aimed to maximize the sum of specificity and sensitivity in differentiating between cases with a clinical diagnosis of drug-induced blood cytopenia and tolerant controls. Sensitivity analysis was conducted using different SPS scores. These statistical analyses were performed using the IBM SPSS Statistics version 21.0 (IBM Corporation, Armonk, NY, United States).

Of the 39 patients (40 cases) with drug-induced blood cytopenia, the most common types were agranulocytosis (n = 29, 72.5%) and neutropenia (n = 6, 15.0%). Less frequent were haemolytic anaemia (n = 2, 5.0%), bicytopenia (n = 2, 5.0%), and pancytopenia (bone marrow aplasia; n = 1, 2.5%). The median [interquartile range, IQR] age was 38 [23.5–59] years, with no significant age differences between cytopenia types. Most patients were female (n = 26, 66.7%). The distribution of cases across age groups was as follows: minors (0–17 years), 22.5% (9/40); adults (18–65 years), 55.0% (22/40); and older adults (>65 years), 22.5% (9/40). Adult patients accounted for the majority of agranulocytosis (58.6%, 17/29) and neutropenia (50%, 3/6) cases.

Patient characteristics are summarized in Table 1. Most cases presented with three or more comorbidities (n = 26, 66.7%) and reported no previous allergies (n = 28, 71.8%). The median [IQR] of comorbidities per patient was 4 [2–6.5]. Most patients with agranulocytosis (16/28, 57.1%), all patients with neutropenia, and all patients with bicytopenia or bone marrow aplasia had three or more comorbidities.

Most patients (87.5%) presented with symptoms, with the majority (72.5%) experiencing two or more symptoms. The median number of symptoms per case was 3. The most common symptoms were fever (80.0%), nausea and vomiting (25.7%), general discomfort (20.0%), and cutaneous lesions (14.3%). Nearly half of the patients (48.7%) experienced concomitant reactions. Complications included non-systemic infections (3 cases), sepsis (1 case), and anaemia (2 cases).

In our study, 72.5% of patients demonstrated unequivocal agranulocytosis, with a median [IQR] neutrophil count at index date of 0.25 × 10³/µL [0.03 × 10³/µL - 0.61 × 10³/µL] at the nadir of neutrophil decline. Thirty percent had neutrophil counts below 0.1 × 10⁹/L, and 15.0% experienced complications, including local infections or sepsis. Among those with neutropenia, 12.5% did not exhibit clinical signs of infection, similar to patients with other forms of cytopenia. Fever was the most common symptom, occurring in 80.0% of cases, while 12.5% remained asymptomatic.

The observed median [IQR] time from drug intake to observation of blood tests alterations (latency) for all cytopenia cases was 8 [3.5–21] days. For those with neutropenia, agranulocytosis, haemolytic anaemia, bicytopenia and the bone marrow aplasia case, the observed median [IQR] latency was 6 [6–12.5], 11 [3–21.25], 4 [4–4], 6.5 [6.25–6.75] and 241 [241–241] days, respectively.

The literature-reported median latency [IQR] (Table 2) for all cytopenia types was 7 [2–15] days. Specifically, the median latency [IQR] was 15 [2–15] days for neutropenia and 7 [2–13.5] days for agranulocytosis. Mean latency data for haemolytic anaemia, bicytopenia, and bone marrow aplasia were unavailable. The median difference [IQR] between observed and literature-reported latency was 0 [-4 to 6] days for all cytopenias, 4 [0–5] days for neutropenia, and 0 [-4 to 6] days for agranulocytosis. No statistically significant differences were found between observed and literature-reported latency.

The median [IQR] time to haematological recovery (defined as WBC count ≥ 3.5 × 10³/µL; ANC ≥ 1.5 × 10³/µL; Hb > 10.0 g/dL; APC ≥ 0.1 × 10⁶/µL) for all cytopenias cases was 7 [5, 16.5] days. Recovery times varied across cytopenia types, ranging from a median of 6 days for agranulocytosis to 90 days for bone marrow aplasia. However, no statistically significant differences in recovery time were observed among the different types of cytopenia (p = 0.201).

The median [IQR] time between the onset of cytopenia and the performance of the LTT was 180.5 [137.75–279.75] days for all cases. This time varied across cytopenia types, ranging from 158 [124–272] days for agranulocytosis to 324 [249–399] days for haemolytic anaemia (Table 2). However, no statistically significant differences were observed among the different types of cytopenia (p = 0.526).

Microbiology tests were performed on 39 of the 40 patients, with 45.0% (18 cases) yielding a positive result (Table 3).

The positive findings revealed a range of viral and bacterial infections across different test types. Serology and PCR tests identified pathogens such as HIV, Epstein-Barr Virus (EBV), and COVID-19. Notably, all four positive PCR results occurred in patients with agranulocytosis. Bacterial infections were also detected, primarily in blood and urine cultures, with Escherichia coli being the most frequently isolated pathogen.

ANCA testing was performed in 15/40 (37.5%) cases. Of these, 2 cases (5.0%) were positive, both of which were agranulocytosis cases. The remaining 13 cases (32.5%) were negative. ANCA testing was conducted in cases of agranulocytosis (11 tests, 2 positive), haemolytic anaemia (2 tests, both negative), and neutropenia (2 tests, both negative). No statistically significant differences in ANCA tests results were found among the different types of cytopenia (p = 0.591) (Table 4).

Treatment was administered in 28/40 (70.0%) cases of cytopenia, most frequently for agranulocytosis (20 cases). Half of all cases (20/40, 50.0%) received granulocyte colony-stimulating factor (G-CSF) or granulocyte-macrophage colony-stimulating factor (GM-CSF) therapy, primarily for agranulocytosis (17 cases), followed by neutropenia (2 cases) and bicytopenia (1 case). Corticosteroids were administered in 15/40 (37.5%) cases, mostly in agranulocytosis (9 cases). Immunosuppressants (cyclosporine) were used in only one case (2.5%), which was a bone marrow aplasia case. Four cases (10.0%) received biological therapy, including normal human immunoglobulin (2 cases), mepolizumab (1 case), and rituximab (1 case). Further details on treatment modalities are provided in Table 5.

The median [IQR] time to resolution for cytopenia cases varied depending on the treatment received, if G-CSF/GM-CSF were administrated was 5 [4.5–7] days, no G/GM-CSF was 18.5 [7–46] days. This difference was statistically significant (p < 0.001), suggesting that treatment with G-CSF or GM-CSF was associated with faster resolution.

Regarding corticosteroids, the median [IQR] time to resolution for cytopenia cases was 7 [5.5–19.50] days when corticosteroids were used and 6 [5–15] days when there were not used. This difference was not statistically significant (p = 0.652).

For biological therapies, two cases were treated with intravenous immunoglobulin (IgIV), one of agranulocytosis and one of haemolytic anaemia (concomitant rituximab use), one case of agranulocytosis was treated with mepolizumab. The resolution time for these cytopenias had a median [IQR] of 7 [6–19] days when biological therapies were used, and 7 [5–16] days when they were not, with no statistically significant differences (p = 0.892). Additionally, one case of pancytopenia was treated with cyclosporine, the time of resolution was 90 days.

The median [IQR] white blood cell (WBC) count at the index date varied across cytopenia types, ranging from 1.28 × 10³/µL [1.09 × 10³/µL, 1.48 × 10³/µL] in bicytopenia cases to 13.98 × 10³/µL [11.79 × 10³/µL, 16.16 × 10³/µL] in haemolytic anaemia cases. However, there were no statistically significant differences in WBC count among the different types of cytopenia (p = 0.060).

In contrast, the absolute neutrophil count (ANC) at the index date showed significant differences among cytopenia types (p = 0.001). The median [IQR] ANC was lowest in agranulocytosis cases (0.27 × 10³/µL [0.04 × 10³/µL, 0.41 × 10³/µL]) and highest in haemolytic anaemia cases (12.20 × 10³/µL [10.13 × 10³/µL, 14.26 × 10³/µL]). Similarly, the absolute platelet count (APC) at the index date differed significantly among cytopenia types (p = 0.045), with the lowest median [IQR] APC observed in bone marrow aplasia (0.01 × 10⁶/µL [0.01 × 10⁶/µL, 0.01 × 10⁶/µL]). Hemoglobin (Hb) levels at the index date also showed significant differences among cytopenia types (p = 0.024). Haemolytic anaemia cases had the lowest median [IQR] Hb level (5.55 mg/dL).

The median [IQR] minimum WBC count varied across cytopenia types, with no statistically significant differences observed (p = 0.209). However, significant differences were found in the minimum ANC (p = 0.006), APC (p = 0.021), and Hb level (p = 0.038) among the different types of cytopenia. Notably, only 2/29 (6.9%) agranulocytosis cases had an ANC <0.01 × 10³/µL.

Eosinophilia was observed in only 2/40 (5.0%) cytopenia cases, both of which were agranulocytosis. There were no statistically significant differences in the presence of eosinophilia (p = 0.939) or the maximum eosinophil count (p = 0.815) among the different types of cytopenia.

Detailed haematological values for each cytopenia case, including WBC counts, are presented in Supplementary Table S2.

The median [IQR] number of drugs studied per patient was 2 [2–3]. In all 40 cytopenia cases, at least one drug had an SPS score +4 or higher, indicating a high level of suspicion for drug causality. Ninety-nine drugs were evaluated, and the majority (n = 90, 90.9%) were suspected to be involved in the blood cytopenia. The median [IQR] SPS score was 6 [5–7]. Of them, 32 (32.3%) had SPS score in the possible range (4–5) category, 54 (54.5%) in the probable (6–7) category, and 4 (4.0%) in the definite (≥8) category. Only 7.1% (n = 7) of drugs had a SPS score below +4, and in two cases, the score could not be determined.

The most frequently evaluated drug by SPS score was metamizole (17.2%), followed by acetaminophen (9.1%) and the combination of amoxicillin and a beta-lactam (8.1%) (Table 6).

The underlying conditions or therapeutic indications for the suspected causative drugs in cases of drug-induced cytopenia are shown in Supplementary Table S3. However, data on the primary pathology of control subjects were not systematically recorded in the source database and were therefore not available for analysis. Nevertheless, all control subjects received the same or similar treatments without developing cytopenias.

In 8 (20.0%) cytopenia cases, only one drug was suspected, while polypharmacy (≥5 drugs) was detected in 2 (5.0%) cases. The distribution of cases by the number of implicated drugs (SPS score ≥+4) is shown in Table 7.

Most drugs implicated in agranulocytosis (92.1%, n = 70/76), haemolytic anaemia (66.7%, n = 2/3), and all drugs implicated in neutropenia, bicytopenia, and bone marrow aplasia had SPS scores ≥+4. No statistically significant differences in SPS score categories (≥+4 vs. <+4) were found among the different types of cytopenia (p = 0.335).

The controls were paired in age with the cases being the median [IQR] age of cases, 38 [23.5–59], and of controls, 49 [32–61] (p = 0.236). There was no statistically significant difference between the proportion of females between the case group (66.7%) and control group (49.4%), (OR: 2.048; 95% CI: 0.929–4.512). The LTT stimulation index (SI) was significantly higher in cases of drug-induced cytopenias compared to tolerant controls (median [IQR]: (1.80 [1.30–2.90]) vs. 1.20 ([1.00–1.50]; p < 0.001) (Figure 1). Conversely, only 1 of 85 (1.2%) controls, who had received rituximab, exhibited a positive LTT (Supplementary Table S1).

To determine the best discriminative threshold for SI, a ROC curve analysis was performed using the 40 cases (in 39 patients) with clinically diagnosed blood cytopenia and 85 tolerant controls. In cases in where multiple drugs were suspected, the maximum SI was used for the analysis. An optimal SI cut-off of 1.95 was identified (Figure 2) yielding a sensitivity of 72% and specificity of 99% (area under the curve [AUC], 0.86; 95% asymptotic confidence interval [CI], 0.77–0.96; p < 0.001).

Using this threshold, 41.4% (41/99) of the suspected drugs resulted in a positive LTT, and at least one LTT was positive for 75.0% (30/40) of cases, indicating a drug-specific immune response mechanism underlying the cytopenia. Regarding the number of positive LTTs per case, 60% (24/40) had only one drug inducing a positive proliferative response, 15% (6/40) had two or more drugs (Table 8).

Analysis of drug involvement, categorized by ATC classification, revealed that in our sample antibacterials (J01) 37.4% (n = 37), analgesics (N02) 26.3% (n = 26) and anti-inflammatory and antirheumatic products (M01) 10.1% (n = 10) were the most frequently implicated drug subgroups in the blood cytopenia cases. A substantial proportion of the implicated drugs (42.4%) showed positive lymphocyte transformation test (LTT) results. Positive LTTs were most frequently associated with Antibacterials (J01, 43.2%), analgesics (N02, 38.4%), immunosuppressants (L04, 25%), anti-inflammatory and antirheumatic products (M01, 20%). These four drug subgroups accounted for 69% of all positive LTTs. Notably, these same four ATC groups also had the highest number of LTTs performed. There were no significant differences in the frequency of positive LTTs among the ATC drug groups (p = 0.397). Table 9 provides a detailed breakdown of the number of drugs, SPS scores, and the number of positive LTTs by ATC group.

A detailed summary of all drugs tested by LTT both in cases and controls, including the number of tests performed in each group for each drug, as well as the absolute terms and percentages of positive and negative results, is provided in Supplementary Table S4.

Among the 41 positive LTTs, 40 (97.6%) had a previous SPS score ≥+4 while only one (2.4%) had a previous SPS score <+4. On the other hand, among negative LTTs, 10.3% (6/58) had a prior SPS score <+4 and 89.7% (52/58) had a score ≥+4. No significant differences in the frequency of positive LTTs depending on the previous SPS score (≥+4 or <+4) were found (p = 0.285). Table 10 lists the specific drugs implicated in each case of blood cytopenia, along with their SPS scores and LTT results.

Positive LTT results were observed in 79.3% (23/29) of agranulocytosis cases, 66.7% (4/6) of neutropenia cases, and all cases of haemolytic anaemia and bone marrow aplasia, while all cases of bicytopenia were negative.

Patients with drugs that elicited a positive LTT result were not intentionally re-challenged with drugs. However, accidental re-exposure occurred in three patients. One of these patients did not tolerate the suspected drug (metamizole) and experienced a recurrence of febrile neutropenia.

On the other hand, for drugs with a final causality score between 4 and 5 (categorized as ‘possible’ in the SPS algorithm) and a negative LTT result, a risk minimization plan with haematological monitoring was recommended if drug re-exposure was considered. Among the 40 cytopenia cases, 23 (57.5%) patients were re-exposed to these drug(s). Notably all re-exposed cases tolerated the drug(s), including 18 (100.0%) cases of agranulocytosis, 3 (100%) cases of neutropenia and 1 (100.0%) case of bicytopenia (Table 11).

A 100% negative predictive value (NPV), with a 95% confidence interval of 94.4%–100%, was observed for a causality score below 6 combined with a negative lymphocyte transformation test (LTT) result. This indicates that in patients with these findings, drug re-exposure under haematological monitoring is likely safe. This high NPV strongly supports the clinical utility of LTT and causality scoring for accurately predicting drug re-exposure tolerance in haematological cytopenias. It enables clinicians to confidently proceed with re-exposure when appropriate, minimizing the risk of adverse reactions, particularly in severe cases like agranulocytosis or febrile neutropenia, and enhancing overall patient safety.