Development of a clinical risk scoring system for artesunate treatment failure prediction in malaria patients

Abstract

Background

Severe malaria remains a major cause of morbidity and mortality worldwide. Early identification of patients at high risk of poor response to treatment is essential, yet no simple clinical tool is currently available. This study aimed to develop a risk scoring system to predict failure to artesunate therapy.

Methods

This retrospective study included adult patients (aged ≥18 years) who received at least one dose of intravenous artesunate at the National Medical Center in South Korea between 2014 and 2023. Treatment failure (early or late clinical failure) was the response variable, which was defined according to WHO criteria. Candidate predictor variables included demographic, clinical, parasitological, and laboratory parameters. Odds ratios (ORs) and adjusted odds ratios (aORs) were calculated using univariate and multivariable logistic regression analyses, respectively. Final predictors were selected through backward elimination based on the Likelihood Ratio criterion, and a clinical risk scoring system was developed based on the adjusted ORs.

Results

Among 98 patients included in the final analysis, treatment failure occurred in 12 (12.2 %). Multivariable analysis identified female sex, parasitemia >5 %, and impaired consciousness as independent risk factors. Using these variables, a risk-scoring system was constructed, and the predicted probabilities of treatment failure for patients with scores of 0, 1, 2, 3, and 4 points were 5 %, 16 %, 41 %, 72 %, and 90 %, respectively (AUROC = 0.768, 95 % CI: 0.605–0.931).

Conclusions

Parasitemia >5 %, impaired consciousness, and female sex were predictive of artesunate treatment failure as defined by WHO clinical criteria. The developed risk scoring system provides a practical tool for identifying high-risk patients requiring intensified monitoring and alternative treatment strategies. These findings are derived from a South Korean cohort and should be interpreted with caution when extrapolated to endemic populations.

Graphical abstract

Highlights

• •A risk score was developed to predict artesunate treatment failure in malaria patients.
• •High parasitemia (>5 %), impaired consciousness, and female sex were key predictors of failure.
• •The tool can guide early identification of high-risk patients needing close monitoring.
• •As This study was conducted in South Korea, the generalisability of the findings to other regions should be considered.

1. Introduction

Malaria remains a significant global health issue, responsible for substantial morbidity and mortality. The recent World Health Organization, 2023 estimated 249 million cases and 608,000 deaths worldwide in 2022, which reflects a persistently high incidence despite ongoing control efforts (Venkatesan, 2024). Although the majority of malaria cases arise in endemic regions, imported infections among travelers and migrants continue to pose a challenge in non-endemic countries (World Health Organization, 2023a). Accordingly, the World Health Organization (WHO) has emphasized the importance of sustained vigilance and rapid response to detect and contain such cases and to prevent re-establishment of transmission. Early recognition and prompt initiation of effective treatment are crucial for reducing mortality and improving clinical outcomes in severe malaria cases.

Intravenous artesunate has demonstrated superior efficacy compared to intravenous quinine for treating severe malaria, significantly reducing mortality by 35 % in Southeast Asian adults and by 22.5 % in African children (SEAQUAMAT, 2005; Dondorp et al., 2010). The WHO recommended intravenous artesunate as the first-line treatment for severe malaria in both adults and children, including cases caused by Plasmodium falciparum, the predominant species responsible for malaria-related mortality, as well as by Plasmodium vivax, and Plasmodium knowlesi (World Health Organization, 2023a). Additionally, several studies conducted in non-endemic countries have shown that artesunate significantly reduced parasite clearance time (PCT), hospital length of stay, intensive care unit (ICU) admissions, and overall complications (Zoller et al., 2011; Eder et al., 2012; Kurth et al., 2015; McCarthy et al., 2015).

Despite evidence supporting its efficacy, treatment failure continues to occur, highlighting enduring challenges in clinical application. In South Korea, a cohort of 55 patients with imported P. falciparum malaria treated between 2014 and 2019 reported an overall mortality of 3.6 % and 10 % among severe cases, with over one-fifth of patients requiring intensive care (Lee et al., 2022). Also, treatment failure rates exceeding 10 % have been reported in some regions (World Health Organization, 2023a, World Health Organization, 2023b), and a meta-analysis indicated that treatment failures could still occur in some patients, even in the absence of drug resistance, with factors such as younger age and higher transmission intensity (Mumtaz et al., 2020). Understanding risk factors associated with artesunate treatment failure is critical for optimizing therapeutic strategies, particularly in high-risk patient populations.

Previous studies have identified a range of clinical and demographic factors potentially associated with artesunate treatment failure, including subtherapeutic dosing, young age, high parasitemia, fever at presentation, and severe malaria (Mishra et al., 2012, 2014; Duong et al., 2021). However, the impact of these factors and their threshold criteria appear to vary across geographic and population contexts. For example, parasitemia >50,000/μL was associated with delayed PCT in African children (Mishra et al., 2014). In contrast, no such association was observed with parasitemia >10 % in Vietnamese adults (Duong et al., 2021), underscoring the lack of a universally applicable threshold and the variability introduced by different measurement units. In Southeast Asia, artemisinin resistance and delayed clearance have been widely reported, whereas artemisinin-based combination therapies remain effective in India and most African regions (Fairhurst, 2015; Woodrow and White, 2017; Arya et al., 2021).

While these regional differences could be attributed to local epidemiological and pharmacological factors, evidence from East Asia is still limited. Also, risk scoring systems, which can facilitate early identification of high-risk patients and guide clinical decision-making, remain undeveloped. This retrospective study aimed to develop a practical risk scoring system for artesunate treatment failure in South Korean patients. Additionally, treatment efficacy and safety were investigated.

2. Materials and methods

2.1. Study population

This retrospective study included adult patients aged 18 years and older who received at least one intravenous dose of artesunate for malaria at the National Medical Center, South Korea, between January 2014 and December 2023. Patients were excluded if subsequent malaria episodes following artesunate treatment could not be clearly classified as reinfection or recrudescence, or if patients who initially tested positive on malaria rapid diagnostic test did not have their infection subsequently confirmed by microscopy or polymerase chain reaction. The study received approval from the Institutional Review Board (IRB) of the National Medical Center (IRB number: 2024-08-072), and the IRB waived the requirement for informed consent. We followed the ethical guidelines outlined in the 1964 Declaration of Helsinki and its subsequent amendments.

2.2. Antimalarial treatment regimens

Intravenous artesunate (Artesune®, Guilin Pharmaceutical Co., Ltd, China/Artesunate Amivas, Amivas Inc., USA) was administered as first-line therapy in accordance with the South Korean national guidelines for severe malaria. The initial dose of 2.4 mg/kg was given at admission, followed by two additional doses at 12-h intervals. If parasitemia remained above 1 % thereafter, the same dose was continued once daily as needed. Once patients were clinically stable and able to take oral medications, treatment was completed with an oral antimalarial regimen such as atovaquone/proguanil or mefloquine, as recommended in the national guidelines.

2.3. Clinical outcome definitions

The primary outcome of this study was treatment failure following artesunate administration, defined according to the WHO guideline for antimalarial drug efficacy (World Health Organization, 2023a). Treatment failure was classified as early treatment failure (ETF, Days 1–3) or late clinical failure (LCF, Days 4–28). ETF was defined as persistence or worsening of parasitemia and clinical signs within the first three days of treatment, including any of the following: parasitemia with severe malaria features, increased parasitemia on Day 2 compared to baseline, fever ≥37.5 °C with parasitemia on Day 3, or parasite levels ≥25 % of baseline after Day 3. LCF was defined as recurrence of parasitemia with danger signs, severe malaria features, or fever ≥37.5 °C between Days 4 and 28 in patients not meeting ETF criteria.

For additional analysis, efficacy was evaluated based on mortality, ICU admission rate, ICU length of stay, and PCT. PCT₅₀, PCT₉₀, and PCT₉₉ were defined as the time required for parasite density to decrease by 50 %, 90 %, and 99 %, respectively, following artesunate administration. PCT was calculated using the parasite clearance estimator algorithm developed by the WorldWide Anti-Malarial Resistance Network (Infectious Diseases Data Observatory IDDO, 2011; Flegg et al., 2011). Parasitemia was measured at 6-h intervals, and the clearance phase was modeled when at least three data points were available within this phase. A lower limit of detection equivalent to approximately 50 parasites/μL was applied, and data from lag or tail phases of the parasitemia profile were excluded to avoid bias in half-life estimation. The safety of artesunate was evaluated based on adverse drug reactions, which were classified according to system organ classes and preferred terms using WHO Adverse Reaction Terminology version 092.

2.4. Data collection

Patient data were retrospectively collected from electronic medical records. Baseline characteristics included sex, age, body weight, body mass index, and underlying medical conditions. Malaria-related information included the use of chemoprophylaxis and the identified Plasmodium species. Treatment-related data included artesunate dose and frequency, and prescribed oral antimalarial medications. Clinical parameters included hospital length of stay, ICU admission, and ICU duration. Laboratory evaluations included hematologic profiles, liver function test, and microscopic examination of blood smears to determine parasite densities (%). Symptom profiles during artesunate treatment, including fever, chills, sweating, headache, myalgia, diarrhea, dizziness, and vomiting, were also documented.

To assess the severity of P. falciparum malaria, we collected data on clinical and laboratory features. Severe malaria was defined using WHO criteria based on clinical features (impaired consciousness, prostration, multiple convulsions, significant bleeding, pulmonary edema and circulatory collapse/shock) and laboratory abnormalities (severe anemia, acidosis, hypoglycemia, acute kidney injury, and jaundice) (World Health Organization, 2023a). For non-falciparum malaria, the same criteria were applied, with the exception that no parasitemia threshold was required. Detailed definitions and criteria are provided in Supplementary Table 1. For impaired consciousness, hypoglycemia was specifically excluded as an alternative cause, and such cases were considered consistent with cerebral malaria.

2.5. Statistical analysis

Statistical analyses were performed according to the characteristics of each variable. Categorical variables were compared using the Chi-squared test or Fisher's exact test, as appropriate. Continuous variables were analyzed using the Mann–Whitney U test due to the small sample size, regardless of the normality assumption, and results were presented as medians with interquartile ranges (IQRs). Adverse drug reaction rates were summarized descriptively, and 95 % confidence intervals (CIs) were calculated using the Clopper–Pearson method. To identify independent predictors of treatment failure, multivariable logistic regression analysis was conducted. Variables with a p < 0.05 in the univariate analysis were initially considered for inclusion in the multivariable logistic model. In addition, age and sex were retained a priori as potential confounders given their established clinical relevance, regardless of statistical significance. Final variable selection in the multivariable model was performed using backward elimination based on the Akaike Information Criterion, with likelihood ratio tests used to evaluate the contribution of variables at each step. To assess the robustness of variable selection, forward and bidirectional stepwise selection were also performed. As a sensitivity analysis, univariate logistic regression analysis was repeated after excluding non-falciparum infections. Odds ratios (ORs) and adjusted odds ratios (aORs), each with 95 % confidence intervals (CIs), were obtained from univariate and multivariable analyses, respectively. A multivariable logistic regression model was used to construct a risk scoring system, in which each variable's aOR was normalized by dividing it by the smallest aOR in the model and rounding to the nearest whole number. The model's performance was evaluated using area under the receiver operating characteristic curve (AUROC) analysis. Observed treatment failure risk was calculated as the proportion of patients with treatment failure within each risk score category, whereas predicted risk was derived from the logistic regression model based on the assigned risk scores. All statistical analyses were performed using R software (4.3.1; R Foundation for Statistical Computing, Vienna, Austria), and statistical significance was defined as a p < 0.05.

3. Results

A total of 115 patients received at least one dose of intravenous artesunate at the National Medical Center between January 2014 and December 2023. After excluding patients in whom reinfection and relapse could not be clearly distinguished (n = 5) and those with an initial positive malaria rapid test followed by a negative confirmatory test (n = 12), 98 patients were included in the final analysis (Fig. 1). Treatment failure was observed in 12 patients (12.2 %), consisting of six cases of early treatment failure (6.1 %) and six cases of late clinical failure (6.1 %). Among these, one patient died, and the remaining 11 achieved clinical recovery following rescue treatment (Supplementary Table 2).

Fig. 1.

Flowchart of study patient selection.

Table 1 presented the baseline characteristics of the enrolled participants. Of the 98 patients, 10 (10.2 %) were female, 9 (9.2 %) were older than 60 years, and 53 (54.1 %) had a body weight greater than 70 kg; none of these demographic factors demonstrated a statistically significant association with treatment failure. The most common comorbidities were hypertension (15.3 %) and diabetes mellitus (8.2 %), but no comorbidity demonstrated a significant difference between the treatment failure and success groups. Additionally, there were no statistically significant differences in the cumulative artesunate dose or in the use of concomitant oral antimalarial agents between the two groups. Among the clinical severity criteria, impaired consciousness (41.7 % vs. 4.7 %, p = 0.001) and jaundice (33.3 % vs. 9.3 %, p = 0.038) were significantly more prevalent in the treatment failure group. Furthermore, patients with parasitemia >5 % on the first day of admission had a significantly higher rate of treatment failure compared to those with parasitemia ≤5 % (50.0 % vs. 10.5 %, p = 0.003).

Table 1.

Association between baseline characteristics and artesunate treatment failure.

Variables	Treatment failure (n = 12)	Treatment success (n = 86)	Unadjusted OR (95 % CI)	p-value
Sex			3.76 (0.84–17.17)	0.103
Female	3 (25.0)	7 (8.1)
Male	9 (75.0)	79 (91.9)
Age (years)			4.44 (0.95–20.89)	0.078
≤ 60	9 (75.0)	80 (93.0)
> 60	3 (25.0)	6 (7.0)
Weight (kg)			0.38 (0.11–1.35)	0.124
≤ 70	8 (66.7)	37 (43.0)
> 70	4 (33.3)	49 (57.0)
BMIa			0.90 (0.22–3.74)	1.000
< 23	3 (27.3)	20 (25.3)
≥ 23	8 (72.7)	59 (74.7)
Comorbidities
Hypertension	3 (25.0)	12 (14.1)	2.06 (0.49–8.69)	0.388
Diabetes mellitus	2 (16.7)	6 (7.0)	2.67 (0.47–15.04)	0.253
Human immunodeficiency virus	1 (8.3)	2 (2.3)	3.82 (0.32–45.66)	0.327
Cancer	0 (0)	3 (3.5)	NA	1.000
Liver disease	0 (0)	3 (3.5)	NA	1.000
Chemoprophylaxis			0.97 (0.84–1.13)	0.572
Yes	0 (0)	7 (8.1)
No	4 (33.3)	18 (20.9)
Unknown	8 (66.7)	61 (70.9)
Plasmodium species			NA	0.721
P. falciparum	12 (100)	71 (82.6)
P. vivax	0 (0)	8 (9.3)
P. ovale	0 (0)	5 (5.8)
P. malariae	0 (0)	2 (2.3)
Artesunate dose (mg/kg)			1.15 (0.36–5.77)	0.747
< 2.4	3 (25.0)	28 (32.6)
≥ 2.4	9 (75.0)	58 (67.4)
Artesunate regimen			0.76 (0.15–3.76)	1.000
Monotherapy	10 (83.3)	68 (79.1)
Co-administration with oral antimalarial drugs	2 (16.7)	18 (20.9)
Manifestation of severity criteria
Impaired consciousness	5 (41.7)	4 (4.7)	14.64 (3.19–67.26)	0.001
Acidosis	1 (8.3)	3 (3.5)	2.52 (0.24–26.35)	0.412
Severe anemia	2 (16.7)	10 (11.6)	1.52 (0.29–7.96)	0.639
Acute kidney injury	5 (41.7)	23 (26.7)	1.96 (0.56–6.78)	0.315
Jaundice	4 (33.3)	8 (9.3)	4.88 (1.20–19.85)	0.038
Circulatory collapse/shock	1 (8.3)	2 (2.3)	3.82 (0.32–45.66)	0.327
Leukocyte count,/μL			0.55 (0.11–2.69)	0.725
< 4000	2 (16.7)	23 (26.7)
≥ 4000	10 (83.3)	63 (73.3)
Platelet,/μL			0.58 (0.11–3.10)	0.621
< 150,000	10 (83.3)	77 (89.5)
≥ 150,000	2 (16.7)	9 (10.5)
AST, U/L			1.95 (0.36–10.50)	0.353
≤ 3 × ULN	10 (83.3)	78 (90.7)
> 3 × ULN	2 (16.7)	8 (9.3)
ALT, U/L			NA	1.000
≤ 3 × ULN	12 (100)	80 (90.3)
> 3 × ULN	0 (0)	6 (7.0)
Parasitemia			8.56 (2.27–32.21)	0.003
≤ 5 %	6 (50.0)	77 (89.5)
> 5 %	6 (50.0)	9 (10.5)

P-values were calculated using Chi-square test or Fisher's exact test. Fisher's exact test was applied when the expected cell counts were less than 5. For comorbidities and manifestations of severity criteria, comparisons were made between patients with and without each condition. For chemoprophylaxis and Plasmodium species, differences were assessed across all categories.

AST, aspartate aminotransferase; ALT; alanine aminotransferase; BMI, body mass index; CI, confidence interval; NA, not available; OR, odds ratio; SBP, systolic blood pressure; ULN, upper normal limit.

^b Patients could meet more than one severity criterion.

^aBMI data were missing for 8 patients.

Variables with p-values <0.05 in univariate analysis, along with age and sex as potential confounders, were included in the multivariable logistic regression model. As shown in Table 2, parasitemia >5 %, impaired consciousness, and female remained in the final model. To assess the robustness of variable selection, forward and bidirectional stepwise procedures were also applied, and all three approaches yielded the same final model. Based on the final multivariable model, a risk scoring system was developed using three variables: female, parasitemia >5 %, and impaired consciousness. Each variable was assigned a point value based on the relative magnitude of its aOR—1 point for female sex, 1 point for parasitemia >5 %, and 2 points for impaired consciousness—resulting in a total score ranging from 0 to 4 (Table 3). The observed treatment failure rates corresponding to scores of 0, 1, 2, 3, and 4 were 5.6 %, 11.8 %, 40.0 %, 75.0 %, and 100.0 %, respectively. Based on the logistic regression model, the predicted probabilities for the same scores were 4.9 %, 15.9 %, 40.9 %, 71.7 %, and 90.3 %, respectively (Fig. 2). The AUROC for the risk score was 0.768 (95 % CI: 0.605–0.931, p < 0.001), indicating acceptable discriminatory performance (Supplementary Fig. 1).

Table 2.

Risk factors for artesunate treatment failure identified through univariate and multivariable analysis, along with a corresponding risk-scoring system.

Variables	Univariate analysis		Multivariable analysis		Score
	Unadjusted OR (95 % CI)	p-value	Adjusted OR (95 % CI)	p-value
Female	3.76 (0.84–17.17)	0.103	5.03 (0.82–30.73)	0.080	1
Age >60 years	4.44 (0.95–20.89)	0.078
Parasitemia >5 %	8.56 (2.27–32.21)	0.003	5.04 (1.08–23.53)	0.040	1
Impaired consciousness	14.64 (3.19–67.26)	0.001	9.90 (1.72–56.89)	0.010	2
Jaundice	4.88 (1.20–19.85)	0.038

Variables were included in the initial multivariable model based on either univariate significance (p < 0.05) or a priori selection as potential confounders (age and sex). Unadjusted ORs are shown for all variables in the initial model. Adjusted ORs are shown for variables retained after backward elimination. Reference groups for each variable were: male, ≤60 years (age), parasitemia ≤5 %, absence of impaired consciousness, and absence of jaundice.

OR, odds ratio; Cl, confidence interval.

Table 3.

The observed artesunate treatment failure (%) and those predicted by the risk-scoring system.

Score	Patients with treatment failure (n)	Total patients (n)	Observed treatment failure risk (%)	Predicted treatment failure risk (%)
0	4	71	5.6	4.9
1	2	17	11.8	15.9
2	2	5	40.0	40.9
3	3	4	75.0	71.7
4	1	1	100.0	90.3

Observed treatment failure risk was calculated as the proportion of patients with treatment failure within each risk score category (patients with treatment failure/total patients × 100). Predicted treatment failure risk was derived from the logistic regression model based on the assigned risk scores.

Fig. 2.

The logistic regression curve of the probability of artesunate treatment failure. The x-axis represents the risk score, and the y-axis indicates the probability of treatment failure (%). Predicted risk probabilities were derived from the final multivariable logistic regression model, and the line connects model-predicted probabilities for each score.

Regarding treatment efficacy (Supplementary Table 3), the median hospital stay for patients treated with artesunate was 5.0 days (IQR: 3.0–8.0 days) and the median ICU stay was 2.5 days (IQR: 1.0–7.3). Among 98 patients, 2 (2.0 %) died during hospitalization, and 18 (18.4 %) required ICU admission. The median times to PCT₅₀, PCT₉₀, and PCT₉₉ were 4.0 h (IQR: 2.8–6.4 h), 8.1 h (IQR: 5.3–13.0 h), and 17.8 h (IQR: 12.7–25.1 h), respectively.

Regarding safety, a total of 87 adverse events were reported during artesunate treatment. Overall, 39 patients (39.8 %) experiencing at least one adverse event, defined as any unfavorable or unintended medical occurrence temporally associated with artesunate administration, regardless of causal relationship. When categorized by system organ class, hematologic adverse events were the most frequently reported, occurring in 20 patients (20.4 %), followed by hepatobiliary disorders in 18 patients (18.4 %) and renal events in 8 patients (8.2 %). The most commonly reported individual adverse reaction was elevated liver enzyme levels (13 patients, 13.3 %), followed by thrombocytopenia (11 patients, 11.2 %) and anemia (9 patients, 9.1 %). The categories (>1 % frequency) are summarized in Supplementary Table 4. All reported adverse events resolved, except in four patients: the two who died, one who voluntarily discontinued treatment, and one who returned to India during treatment.

In the sensitivity analysis restricted to falciparum patients, univariate analysis showed that female sex, parasitemia >5 %, impaired consciousness, and jaundice were significantly associated with treatment failure, consistent with the findings observed across all species. The corresponding ORs (95 % CI) were 5.58 (1.07–29.10), 6.89 (1.82–26.05), 16.19 (3.17–82.56), and 4.57 (1.09–19.13), respectively.

4. Discussion

In this study, treatment failure was observed in 12 of 98 patients (12.2 %) who received intravenous artesunate for malaria treatment. After controlling for confounding factors, high parasitemia exceeding 5 % and impaired consciousness at admission were significantly associated with artesunate treatment failure. Additionally, the risk-scoring system demonstrated that higher scores correlated with an increased probability of treatment failure.

The overall treatment failure rate of 12.2 % in our study was lower than the 34.5 % reported by Duong et al. in patients with severe malaria (Duong et al., 2021), but similar to the approximately 10 % failure rate reported in meta-analyses of uncomplicated P. falciparum malaria (Sinclair et al., 2009). This finding may be attributed to the inclusion of both severe and non-severe malaria patients in our study, resulting in a relatively low proportion of severe cases (approximately 36 %). While artesunate is the recommended first-line treatment for severe malaria, in South Korea, empirical treatment is often administered regardless of severity based on comprehensive evaluation of rapid diagnostic test results, clinical symptoms, and travel history.

Impaired consciousness at admission showed a significant association with artesunate treatment failure in our study. Duong et al. reported that severe malaria diagnosis was associated with early treatment failure, with the most commonly observed severe manifestations being jaundice (24.1 %), impaired consciousness (20.7 %), and shock (8.6 %) (Duong et al., 2021). Similarly, Rovira-Vallbona et al. reported that among six patients with treatment failure following artesunate combination therapy for imported malaria, three presented with impaired consciousness at admission (Rovira-Vallbona et al., 2019). Additionally, impaired consciousness reflects cerebral malaria, characterized by blood–brain barrier disruption and microvascular obstruction, which may plausibly impair drug delivery to brain tissue and contribute to poorer treatment outcomes (Trivedi and Chakravarty, 2022).

Patients with parasitemia >5 % also had a higher risk of artesunate treatment failure in our study. The mechanistic basis for this association may involve enhanced inflammatory cytokine responses triggered by higher parasite loads, which can exacerbate tissue damage and impair clinical recovery (Ademolue et al., 2017). However, the thresholds for risk appear to vary across different geographic regions. Rovira-Vallbona et al. reported that 83 % of patients with treatment failure after artesunate combination therapy had parasitemia of approximately 4 % or higher in Belgium (Rovira-Vallbona et al., 2019), while Landre et al. reported a clinical case of treatment failure in a patient with parasitemia of approximately 13 % in France (Landre et al., 2021). Conversely, Abuaku et al. found no association between parasitemia ≥5 % and artesunate treatment failure in pediatric patients in Ghana (Abuaku et al., 2021). These differences likely reflect regional epidemiology, patient demographics, immune status, and clinical settings. In this context, our findings suggested that parasitemia >5 % could serve as a practical criterion for risk scoring and clinical decision-making in East Asian populations.

In our study, although female sex did not reach statistical significance as an independent predictor, it remained in the final risk scoring system, suggesting potential explanatory value. Experimental data supported this possibility, showing that artesunate induced enhanced insulin responses and antioxidant enzyme activation in males, but not in females (Alagbonsi et al., 2021). These pharmacological differences may interact with estrogen-mediated immune responses, which are known to enhance protection during natural malaria infection (Wu et al., 2025). However, such heightened immune activation could paradoxically become detrimental during artesunate therapy, potentially contributing to excessive inflammatory responses and poorer treatment outcomes in female patients.

The mortality rate in our study was consistent with previous reports. For instance, among severe malaria patients treated with intravenous artesunate reported mortality rates of 1.8 % from April 2019 to December 2020 (Abanyie et al., 2021). Our findings regarding ICU admission rates and hospital/ICU length of stay are also consistent with previous studies (Eder et al., 2012; Khuu et al., 2017; Roussel et al., 2021; Botta et al., 2022). Specifically, the median length of hospital stay in previous studies ranged from 4 to 6 days, comparable to 5 days in our study. Reported ICU admission rates varied between 12 % and 27 %, with our study falling within this range at 18 %. Additionally, the median duration of ICU hospitalization in prior research was between 1 and 2 days, while our study observed a slightly longer median of 2.5 days.

Regarding PCT, a study in Europe reported PCT₅₀ of 4.4 h (95 % CI: 3.9–5.2), PCT₉₀ of 14.8 h (95 % CI: 13.0–17.2), and PCT₉₉ of 29.5 h (95 % CI: 25.9–34.4) (Kurth et al., 2015). Additionally, in Netherlands and Belgium, 185 patients with P. falciparum malaria reported PCT₉₉ of 36 h (IQR: 24–48) (Bruneel et al., 2010). Our PCT values were generally shorter than those in previous studies, suggesting that artesunate could rapidly clear parasites in South Korean malaria patients.

Intravenous artesunate is known to be a relatively safe drug with good tolerability across diverse populations (Kreeftmeijer-Vegter et al., 2012; Kurth et al., 2015; Abanyie et al., 2021). A large safety study involving over 1100 patients treated with intravenous artesunate reported relatively low rates of common adverse drug reactions (Ampadu et al., 2018). The most frequently reported adverse reactions were fever (3.5 %), abdominal pain (2.5 %), diarrhea (1.7 %), and asthenia (1.5 %). A review article reported that non-hematological adverse reactions to artesunate were rare and mainly included mild hepatitis, neurological, renal, dermatological, and cardiac symptoms (Roussel et al., 2017). In our study, no serious adverse reactions to intravenous artesunate were reported, with hematological adverse events including anemia being the most common (20.4 %). These findings support the overall favorable safety profile of intravenous artesunate.

This study has several limitations. First, as a single-center retrospective study, the number of patients was limited. Second, although our risk scoring system demonstrated good discriminatory performance, external validation in different populations and ethnic groups is necessary before widespread clinical implementation. Third, the findings may not be directly generalisable to populations permanently residing in high transmission settings, because our study population mainly consisted of returned travelers or migrant workers, and this epidemiological feature also explains the male predominance observed in our cohort. Similar patterns have been consistently reported in other South Korean cohorts (Lee et al., 2022), supporting that this reflects the travel-related nature of malaria in non-endemic settings rather than study-specific bias. Fourth, treatment failure was defined according to WHO clinical criteria (ETF and LCF), which reflect clinical and parasitological non-response but do not confirm artemisinin resistance; molecular markers and pharmacokinetic data would be required for such confirmation. Also, the small number of ETF and LCF cases limited separate analysis of these distinct treatment failure outcomes. Finally, patients were included if they received at least one dose of intravenous artesunate, which may have encompassed incomplete treatment courses and could influence outcome interpretation. Future studies should therefore include prospective, multicenter designs with larger and more diverse patient populations, incorporate molecular and pharmacokinetic assessments, and investigate the biological mechanisms underlying sex-related differences to validate and refine the proposed risk scoring system.

Despite these limitations, this study represents the first analysis of risk factors for treatment failure in patients receiving intravenous artesunate in South Korea, providing valuable insights within a non-endemic, travel-related malaria context. The risk scoring system developed based on these factors offers a simple, clinically applicable tool for risk stratification.

CRediT authorship contribution statement

Dasom Kim: Writing – review & editing, Writing – original draft, Software, Resources, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Da Hoon Lee: Writing – review & editing, Writing – original draft, Validation, Software, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Yubin Song: Writing – review & editing, Funding acquisition, Data curation, Conceptualization. Jung Sun Kim: Writing – review & editing, Supervision, Project administration, Methodology, Data curation, Conceptualization. Hye Sun Gwak: Writing – review & editing, Supervision, Project administration, Funding acquisition, Conceptualization.

Data statement

The metadata supporting the findings of this study have been provided in the supplementary data with all personal identifiers removed. In addition, the R code used for the statistical analyses is available in the supplementary materials.

Funding

This study was supported by a grant from the National Medical Center, South Korea.

Declaration of interest

The all authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article: