Inflammation and Routine Laboratory Associations in Imported Falciparum Malaria Assessed by Partial Correlation Analysis

Abstract

Purpose

To examine whether associations among routine admission laboratory measurements in imported Plasmodium falciparum malaria persist after accounting for other measured variables, including C-reactive protein (CRP).

Methods

This retrospective single-centre study included all adults recorded in a departmental malaria database as hospitalised with microscopically confirmed P. falciparum monoinfection between 2010 and 2025. A rank-based precision matrix yielded 28 partial correlations among seven laboratory measurements and age, each adjusted for the other six variables. Bonferroni correction was applied to the primary matrix. CRP-adjusted subset analyses and Firth logistic models were exploratory.

Results

The primary cohort comprised 48 patients. Three partial correlations survived correction: alanine aminotransferase with aspartate aminotransferase (partial rho +0.88), haemoglobin with red blood cell count (partial rho +0.82) and CRP with platelet count (partial rho − 0.56). Platelet associations with procalcitonin and urea were attenuated after CRP adjustment (n = 39 and 35). In univariate Firth models, CRP was associated with concurrent severe thrombocytopenia (15 events; odds ratio 3.57 per standard deviation, 95% CI 1.70 to 9.15), whereas evidence for an association with parasitaemia was inconclusive.

Conclusion

Routine laboratory abnormalities in falciparum malaria should be interpreted in their inflammatory and clinical context; their co-occurrence does not establish a direct link between organ injuries. Partial correlation analysis can explore this distinction but cannot determine whether an association is caused by inflammation or is organ-specific. These findings support contextual interpretation, not new diagnostic thresholds or treatment recommendations, and require prospective external validation.

Introduction

Malaria remains a major global health problem, with an estimated 282 million cases and 610,000 deaths in 2024 [1]. Imported infections also require clinical expertise in non-endemic settings: 6,131 confirmed malaria cases were reported in the European Union and European Economic Area in 2022, and 99.8% of cases with known importation status were travel-related [2]. In Plasmodium falciparum malaria, parasite sequestration, endothelial activation and the host inflammatory response can affect several organ systems. Experimental human infection studies demonstrate that endothelial activation can precede detectable microvascular dysfunction, illustrating that biological responses and clinically evident organ injury need not occur simultaneously [3].

Routine laboratory abnormalities provide complementary but overlapping information. Thrombocytopenia and increased C-reactive protein (CRP) are common in malaria, although neither is specific to the infection [4, 5]. Human studies document platelet binding to infected erythrocytes and killing of intraerythrocytic parasites [6], while observational studies show that aminotransferase abnormalities may evolve during and after treatment [7, 8]. Longitudinal measurements also demonstrate changes in hepcidin, ferritin and erythropoietic markers during experimental and clinical malaria [9]. These observations caution against assigning each abnormal laboratory value to a single organ or mechanism.

Recent studies of imported malaria have examined CRP in combination with albumin or endothelial biomarkers to assess disease severity and complications [10, 11]. Such work concerns clinical risk assessment. A complementary question is whether correlations between routine measurements persist after adjustment for other measured variables. A low platelet count associated with an elevated renal or inflammatory marker, for example, need not indicate a direct pathological connection between those processes.

Precision matrices summarise partial correlations within a set of variables and have been used to reconstruct metabolic reaction networks from serum metabolomics data [12] and, in the behavioural sciences, to estimate psychological networks from questionnaire data [13]. Their interpretation depends on model assumptions, the variables included and estimation uncertainty; a conditional association does not establish causality [13, 14]. We therefore used a rank-based precision matrix to examine routine admission laboratory associations in adults with imported P. falciparum monoinfection. Our aim was to identify associations that persisted after mutual adjustment and to explore the contribution of CRP-related variation, without treating the resulting network as proof of distinct causal pathways.

Material and Methods

Study Design and Population

This retrospective, single-centre observational study was conducted at a tertiary referral centre for tropical infections. Patients were identified from a departmental database of malaria admissions compiled from the hospital discharge summaries. All adults recorded there as admitted between 2010 and 2025 with imported, microscopically confirmed P. falciparum monoinfection were eligible. Diagnosis rested on Giemsa-stained thick and thin blood films read by experienced microscopists, with species identified morphologically. Patients with mixed Plasmodium infections, non-falciparum species, unspecified species or incomplete species documentation were excluded from the primary analysis. Demographic and clinical data, including sex, age, country of travel, chemoprophylaxis history and clinical signs, were extracted from the electronic medical records.

Data Collection and Laboratory Parameters

Admission laboratory data were extracted from electronic medical records. Seven routine laboratory measurements were selected for the primary analysis: haemoglobin (g/dL), red blood cell count (× 10⁶/µL), platelet count (× 103/µL), alanine aminotransferase (U/L), aspartate aminotransferase (U/L), CRP (mg/L) and serum creatinine (mg/dL). Age (years) was included as a demographic covariate, giving eight primary variables. Procalcitonin (ng/mL), serum urea (mg/dL) and peripheral parasitaemia (%) were used in secondary analyses where available. All values were the first measurements at admission, before antimalarial therapy.

Statistical Analysis

Analyses were performed in Python 3.12 with NumPy, SciPy and pandas. Two-tailed p-values below 0.05 were considered nominally significant. Continuous variables were summarised as medians with interquartile ranges (25th to 75th percentile), and categorical variables as counts and proportions. Thrombocytopenia was defined as a platelet count below 150 × 103/µL and severe thrombocytopenia as below 50 × 103/µL. Alanine aminotransferase above 56 U/L and creatinine above 1.5 mg/dL were used as descriptive thresholds for elevated admission values. A single creatinine threshold was not used to diagnose acute kidney injury, which requires assessment of changes in creatinine or urine output [15]. Multiplicity was controlled by Bonferroni correction within the primary precision matrix only; all subsequent analyses were exploratory, with nominal, uncorrected p-values.

Precision matrix and partial correlations. All eight primary variables were rank-transformed, and the Spearman correlation matrix was computed for patients with complete data for these variables. The inverse matrix was used to estimate pairwise partial correlations, each controlling for the six remaining variables. This quantifies linear associations between rank residuals and is not, without additional distributional assumptions, an exact Gaussian graphical model. In a Gaussian graphical model, zero partial correlation corresponds to conditional independence [13]; we do not assume that equivalence here, because the variables were rank-transformed and no distributional model was fitted. We therefore interpret the coefficients as adjusted associations rather than direct biological links. The matrix was inverted directly without regularisation. Significance was assessed using the approximate t-statistic for partial correlations, with n − 8 degrees of freedom [16]. This parametric approximation should be interpreted cautiously with rank-based estimates and a small sample. The Bonferroni threshold was 0.05/28, approximately 0.0018.

Secondary analyses. Non-parametric bootstrap 95% confidence intervals (10,000 resamples, random seed 42, percentile method) were computed for the bivariate Spearman correlation of every pair whose bivariate correlation met the Bonferroni threshold [17]. These intervals describe bivariate, not partial, correlations. Selected pairwise correlations were also recomputed after adjustment for CRP alone by regressing the rank-transformed variables on rank-transformed CRP and taking the Spearman correlation of the residuals; p-values for these single-covariate partial correlations were obtained from the t approximation with n − 3 degrees of freedom. These analyses used available subsets and assessed attenuation of associations, without presuming that CRP was a causal confounder. Multiple linear regressions provided descriptive consistency checks. Between-group effect sizes were estimated using the rank-biserial correlation derived from the Mann–Whitney U statistic [18]. Univariate Firth penalised logistic models examined associations with severe thrombocytopenia at admission [19]. Predictors were standardised, so odds ratios represent a one-standard-deviation increase; 95% confidence intervals were profile penalised-likelihood intervals. Discrimination was described by the apparent in-sample c-statistic, without optimism correction or cross-validation. The models describe concurrent laboratory findings rather than prediction of a future event.

Results

Study Population

Between 2010 and 2025, 52 adults were admitted with imported, microscopically confirmed P. falciparum monoinfection. 48 of them had complete admission data for the eight primary variables and constituted the analysis cohort. The cohort was predominantly male (35 of 48, 72.9%), with a median age of 43.5 years (interquartile range 32.0 to 49.3; range 23.0 to 72.0). All 48 patients had complete data for the eight primary variables; procalcitonin, urea and parasitaemia were available for 39, 35 and 43 patients, respectively. One patient with gametocytes only on the admission blood film was recorded as 0% asexual parasitaemia. Median admission values are shown in Table 1. Thrombocytopenia was present in 41 of 48 patients (85.4%), including severe thrombocytopenia in 15 of 48 (31.3%). Alanine aminotransferase exceeded 56 U/L in 20 of 48 patients (41.7%), and creatinine exceeded 1.5 mg/dL in 5 of 48 (10.4%). These are descriptive biochemical thresholds and do not establish organ failure or acute kidney injury.

Table 1.

Admission characteristics of the 48 adults with imported P. falciparum monoinfection

Parameter	n	Median (IQR)	Range
Age (years)	48	43.5 (32.0 to 49.3)	23.0 to 72.0
Haemoglobin (g/dL)	48	13.9 (12.3 to 15.0)	8.2 to 17.2
Red blood cells (× 10⁶/µL)	48	4.78 (4.09 to 5.07)	2.56 to 5.63
Platelets (× 103/µL)	48	61 (43 to 128)	14 to 436
Alanine aminotransferase (U/L)	48	46 (25 to 79)	15 to 420
Aspartate aminotransferase (U/L)	48	40 (28 to 78)	13 to 541
C-reactive protein (mg/L)	48	124.1 (68.6 to 189.8)	7.6 to 367.1
Procalcitonin (ng/mL)	39	2.51 (0.87 to 5.71)	0.03 to 100.00
Serum urea (mg/dL)	35	30 (23 to 42)	4 to 161
Creatinine (mg/dL)	48	1.02 (0.87 to 1.21)	0.48 to 3.43
Parasitaemia (%)	43	1.40 (0.35 to 3.00)	0.00 to 16.00

IQR interquartile range. All 48 patients had complete data for the eight primary analysis variables (age, haemoglobin, red blood cell count, platelet count, alanine aminotransferase, aspartate aminotransferase, C-reactive protein and creatinine). Parasitaemia refers to asexual forms; one patient with gametocytes only on the admission film was recorded as 0%

Patients had acquired their infection in 20 countries, almost exclusively in sub-Saharan Africa (Western Africa 25 patients, Middle Africa 13, Eastern Africa 8); a single patient returned from India (Fig. 1). The most frequent countries of exposure were Cameroon (8 patients), Nigeria (6) and Guinea (4). The country of exposure was not recorded for one patient.

Fig. 1.

Geographic origins of the imported P. falciparum cases, for the 47 of 48 patients with a recorded country of exposure. Each patient was assigned the primary (first-listed) country of exposure; several patients reported multi-country itineraries, and regions follow the United Nations geoscheme. Colour intensity reflects the number of patients per country, numbers indicate patient counts, and countries with no cases are shown in grey. Some countries in West Africa are shaded but left unlabelled to reduce clutter

Precision Matrix Analysis and Bivariate confidence Intervals

The primary correlation matrix had a condition number of 39.4. Of the 28 pairwise partial correlations, each controlling for six remaining variables (40 degrees of freedom), three survived Bonferroni correction (Table 2, Fig. 2). These were alanine aminotransferase with aspartate aminotransferase (partial rho +0.88, t = 11.74, p < 0.001), haemoglobin with red blood cell count (partial rho +0.82, t = 8.98, p < 0.001) and CRP with platelet count (partial rho −0.56, t = −4.30, p < 0.001). We refer to these pairs descriptively as haematological, inflammatory-thrombocytopenic and hepatocellular axes; the labels do not establish mechanistic independence or organ specificity. One further pair reached nominal significance only: CRP with creatinine (partial rho +0.34, p = 0.027). The remaining 24 partial correlations did not reach nominal significance. Bootstrap 95% confidence intervals excluded zero for every pair whose bivariate correlation met the Bonferroni threshold: haemoglobin with red blood cell count, rho +0.84 (+ 0.68 to +0.94), platelet count with alanine aminotransferase, rho −0.46 (− 0.70 to −0.16), platelet count with aspartate aminotransferase, rho −0.52 (− 0.74 to −0.24), CRP with platelet count, rho −0.70 (− 0.83 to −0.51), alanine aminotransferase with aspartate aminotransferase, rho +0.91 (+ 0.83 to +0.95), aspartate aminotransferase with CRP, rho +0.47 (+ 0.20 to +0.68) and CRP with creatinine, rho +0.49 (+ 0.21 to +0.72). These intervals do not measure the stability of the partial-correlation network.

Table 2.

Partial correlations reaching nominal significance in the primary matrix (n = 48)

Variable pair	Simple rho	Partial rho	t	p	Axis
Alanine aminotransferase and aspartate aminotransferase	+0.91	+0.88	+11.74	<0.001	Hepatocellular
Haemoglobin and red blood cell count	+0.84	+0.82	+8.98	<0.001	Haematological
C-reactive protein and platelet count	−0.70	−0.56	−4.30	<0.001	Inflammatory-thrombocytopenic
C-reactive protein and creatinine	+0.49	+0.34	+2.30	0.027	Nominal only

The first three pairs survived Bonferroni correction (p < 0.0018); the remaining one reached nominal significance only. The other 24 of 28 partial correlations were non-significant. Degrees of freedom = 40

Fig. 2.

Partial correlation network of seven admission laboratory measurements and age in imported P. falciparum malaria (n = 48). Edges represent partial correlations adjusted for the six other variables. Green edges are positive and red edges are negative; edge width reflects correlation magnitude. Solid edges survived Bonferroni correction and are labelled as three descriptive axes. The dashed edge (CRP-creatinine) reached nominal significance only. Absence of an edge does not demonstrate absence of a biological relationship, and the axis labels do not establish causal or organ-specific mechanisms

Associations After Adjustment for C-Reactive Protein

In exploratory subset analyses (Table 3), two associations were attenuated after CRP adjustment: the platelet-procalcitonin association (n = 39; bivariate rho −0.54, p < 0.001; adjusted partial rho −0.25, p = 0.133) and the platelet-urea association (n = 35; bivariate rho −0.65, p < 0.001; adjusted partial rho −0.28, p = 0.107). The alanine-aspartate aminotransferase and haemoglobin-red blood cell associations changed little in magnitude (− 0.05 and 0.00, respectively). The urea-creatinine association weakened but remained nominally significant (n = 35; bivariate rho +0.77, p < 0.001; adjusted partial rho +0.53, p = 0.001). The platelet-aspartate aminotransferase association weakened but remained nominally significant (n = 48; bivariate rho −0.52, p < 0.001; adjusted partial rho −0.36, p = 0.013). The platelet-alanine aminotransferase association weakened (n = 48; bivariate rho −0.46, p = 0.001; adjusted partial rho −0.29, p = 0.051). Parasitaemia was correlated with platelet count before adjustment (n = 43; rho −0.37, p = 0.014); after adjustment for CRP the association no longer reached nominal significance (adjusted partial rho −0.21, p = 0.186). Attenuation is compatible with shared CRP-related variation, but the change from a significant to a non-significant p-value does not itself establish confounding or a statistically significant difference between coefficients.

Table 3.

Bivariate versus C-reactive-protein-adjusted partial correlations for selected pairs

Variable pair	Simple rho (p)	Adjusted partial rho (p)	Change	Interpretation
Platelet count and procalcitonin	−0.54 (< 0.001)	−0.25 (0.133)	−0.29	Attenuated after CRP
Platelet count and urea	−0.65 (< 0.001)	−0.28 (0.107)	−0.37	Attenuated after CRP
Platelet count and parasitaemia	−0.37 (0.014)	−0.21 (0.186)	−0.16	Weakened; nominal p > 0.05
Urea and creatinine	+0.77 (< 0.001)	+0.53 (0.001)	−0.24	Weakened; nominal p < 0.05
Platelet count and aspartate aminotransferase	−0.52 (< 0.001)	−0.36 (0.013)	−0.16	Weakened; nominal p < 0.05
Platelet count and alanine aminotransferase	−0.46 (0.001)	−0.29 (0.051)	−0.17	Weakened; nominal p > 0.05
Alanine aminotransferase and aspartate aminotransferase	+0.91 (< 0.001)	+0.86 (< 0.001)	−0.05	Little change after CRP
Haemoglobin and red blood cell count	+0.84 (< 0.001)	+0.84 (< 0.001)	0.00	Little change after CRP

Change is the difference in absolute correlation magnitude after adjustment for CRP. p-values for adjusted partial correlations use the t approximation with n − 3 degrees of freedom. All p-values are nominal; these analyses are exploratory. CRP C-reactive protein

Multiple Regression Consistency Checks

Multiple regression provided consistency checks using the same observations and therefore did not independently validate the correlation findings. Alanine aminotransferase was associated with aspartate aminotransferase (coefficient 1.01, t = 16.54, p < 0.001) in a model also including CRP (p = 0.589), platelet count (p = 0.138) and age (p = 0.190); adjusted R-squared was 0.88 (F = 89.52, p < 0.001, n = 48). CRP was associated with platelet count (coefficient −0.539, t = −3.07, p = 0.004) in a model including procalcitonin (p = 0.384), aspartate aminotransferase (p = 0.381), creatinine (p = 0.727) and age (p = 0.390); adjusted R-squared was 0.32 (F = 4.53, p = 0.003, n = 39). In a third model with creatinine as the dependent variable and urea, CRP, platelet count, age and haemoglobin as predictors (n = 35; adjusted R-squared 0.65, F = 13.79, p < 0.001), creatinine was associated with urea (t = 4.17, p < 0.001) and CRP (t = 2.22, p = 0.034); platelet count (t = 0.88, p = 0.384), age (t = 0.69, p = 0.494) and haemoglobin (t = 0.29, p = 0.772) did not reach nominal significance. These exploratory results do not establish the mechanism underlying creatinine variation.

Effect Sizes and Associations with Severe Thrombocytopenia

Compared with patients with normal platelet counts, those with thrombocytopenia (n = 41) had large rank-biserial effect estimates for CRP and procalcitonin (r = 0.87 and 0.86). Only seven patients had normal platelet counts, of whom five had a procalcitonin measurement and five had a urea measurement; these effect estimates are therefore imprecise. In univariate Firth models of concurrent severe thrombocytopenia (15 events in the primary cohort; Table 4), the odds ratio for CRP was 3.57 per standard deviation (95% confidence interval 1.70 to 9.15; apparent c-statistic 0.81). Urea (odds ratio 26.27, 3.09 to 1061), creatinine (odds ratio 3.92, 1.54 to 15.88), age (odds ratio 2.12, 1.12 to 4.47) and aspartate aminotransferase (odds ratio 2.19, 1.08 to 5.99) were also nominally associated. Evidence for an association with parasitaemia was inconclusive (1.16, 0.64 to 2.10). Intervals for urea span more than two orders of magnitude and indicate near-separation rather than a precisely estimated effect. The number of patients and events in each model is given in Table 4. These separate univariate models do not establish independent effects. Their apparent c-statistics are descriptive and do not quantify performance in new patients.

Table 4.

Univariate Firth models of concurrent severe thrombocytopenia (platelet count below 50 × 10³/µL)

Predictor (per SD)	n (events)	Odds ratio (95% CI)	c-statistic	p
Serum urea	35 (12)	26.27 (3.09 to 1061)	0.89	<0.001
C-reactive protein	48 (15)	3.57 (1.70 to 9.15)	0.81	<0.001
Creatinine	48 (15)	3.92 (1.54 to 15.88)	0.72	0.001
Age	48 (15)	2.12 (1.12 to 4.47)	0.72	0.020
Aspartate aminotransferase	48 (15)	2.19 (1.08 to 5.99)	0.67	0.026
Procalcitonin	39 (14)	1.78 (0.95 to 5.77)	0.78	0.077
Alanine aminotransferase	48 (15)	1.61 (0.93 to 3.32)	0.67	0.090
Parasitaemia	43 (15)	1.16 (0.64 to 2.10)	0.64	0.602

Odds ratios are per one standard deviation increase; 95% CIs are profile penalised-likelihood intervals. Models differ in the number of patients with the predictor available, shown as n (events). The c-statistic is apparent in-sample discrimination without optimism correction. All p-values are nominal and exploratory. CI confidence interval; SD standard deviation

Discussion

In this cohort, three prominent laboratory associations persisted after adjustment for the other measured variables, while selected platelet associations with procalcitonin and urea weakened after CRP adjustment. The contribution of the study is to distinguish unadjusted co-variation from associations that persist within a specified adjustment model. It does not establish that systemic inflammation dominates malaria pathogenesis, or that the retained pairs represent biologically independent organ pathways. Here, an organ-related association denotes co-variation between measurements commonly used to assess an organ system, rather than a mechanism specific to that organ.

The inverse CRP-platelet association is consistent with recent observations that inflammatory and haematological markers provide overlapping information in malaria. Nishimura et al. [5] reported increased CRP and lower platelet counts in malaria-positive febrile patients, while Gebreweld et al. [4] documented frequent thrombocytopenia in acute malaria. Both studies were performed in endemic settings and included predominantly P. vivax cases, so they provide contextual support rather than direct validation of our findings in imported falciparum malaria. In adults with imported falciparum malaria, Hoffmeister and Aguilar Valdez [10] reported that the CRP-to-albumin ratio was associated with severity and respiratory complications. Balerdi-Sarasola et al. [11] studied CRP together with angiopoietin-2 for identifying severe imported malaria, and a subsequent cohort study related severe thrombocytopenia and elevated CRP to life-threatening conditions [20]. These studies support evaluating routine markers within a broader clinical assessment, but their outcomes differ from our endpoint of concurrent severe thrombocytopenia.

The present data cannot identify the mechanism connecting CRP and platelet count. Kho et al. [6] demonstrated platelet binding to infected erythrocytes and platelet-associated parasite killing in human malaria, highlighting biological roles beyond a passive response to inflammation. Experimental infection studies also show that endothelial activation can occur early [3]. Our analysis measured neither platelet activation nor complement deposition, cytokines or endothelial biomarkers. It therefore cannot assign platelet depletion to CRP-mediated complement activation or exclude other pathways. Likewise, the inconclusive parasitaemia association does not establish that inflammation is more important than parasite burden. Peripheral parasitaemia and a single CRP measurement provide incomplete representations of processes that vary over the course of infection.

The aminotransferase association persisted after adjustment, but this does not demonstrate that hepatic injury is independent of inflammation. Reuling et al. [8], studying imported uncomplicated falciparum malaria and controlled human malaria infection, found transient liver-test abnormalities that peaked in the days after antimalarial treatment was started and resolved within weeks, and that were associated with inflammatory markers rather than with the antimalarial regimen used. Woodford et al. [7] showed that aminotransferase elevations can peak after diagnosis, so admission measurements capture only one phase of the trajectory. These observations are compatible with coordinated aminotransferase variation that a single CRP value does not fully explain. The observed aspartate-to-alanine aminotransferase ratio in our cohort (median 1.02, interquartile range 0.82 to 1.27) is non-specific and cannot establish an ischaemic or other particular injury mechanism. Persisting co-variation between alanine and aspartate aminotransferase supports retaining an organ-focused interpretation of abnormal liver tests alongside inflammatory context, without using the correlation to diagnose hepatic dysfunction.

The strong haemoglobin-red blood cell association is also biologically expected. Its persistence after CRP adjustment cannot distinguish haemolysis, erythrocyte clearance, impaired production or dilution, and it does not exclude inflammation-related effects on erythropoiesis. Woolley et al. [9] documented longitudinal changes in hepcidin, ferritin and soluble transferrin receptor in experimental and clinical malaria, including abnormalities persisting after the acute infection. A single haemoglobin and red blood cell measurement therefore cannot determine which mechanisms are active. Reticulocyte counts, haemolysis markers and iron-status markers, combined with repeated measurements, would be needed to examine those mechanisms.

Attenuation of the platelet-procalcitonin and platelet-urea associations after CRP adjustment illustrates why an unadjusted laboratory correlation should not automatically be interpreted as a link between separate organ injuries. However, CRP is a marker of the host response and is not established here as a causal confounder. Depending on the underlying causal structure, adjustment may reduce confounding, remove part of a mediated association or introduce bias; the appropriate interpretation requires information beyond changes in correlation coefficients [21]. In a paediatric falciparum study excluding overt bacterial infection, Katte et al. [22] found elevated procalcitonin without clear discrimination between severe and uncomplicated disease. Although that population differs from ours, it reinforces the need for clinical context when interpreting procalcitonin. Our attenuation analysis cannot diagnose or exclude bacterial coinfection and does not justify antibiotic decisions.

The renal findings require particular restraint. Few patients exceeded the descriptive creatinine threshold, and urea was available only in a subset. A CRP-creatinine association reached nominal significance (partial rho +0.34, p = 0.027) but did not survive correction. Neither observation shows that renal abnormalities are simply a consequence of systemic inflammation, and the absence of a renal edge surviving correction can equally reflect limited power, restricted variation or an incomplete set of variables. Moreover, a single admission creatinine above 1.5 mg/dL does not establish acute kidney injury. Assessment requires baseline or serial creatinine and/or urine output in the appropriate clinical setting [15]. Renal assessment should therefore not be scaled back on the basis of this network.

For physicians, the practical message is to interpret platelet count, inflammatory markers and organ-related tests together with the patient’s clinical status and their evolution over time. Co-occurring abnormalities should prompt assessment rather than an assumption of a direct inter-organ mechanism. For laboratory diagnosticians, correlation between abnormal results should not be presented as evidence of the anatomical source or cause of injury; interpretation should incorporate inflammatory context, relevant organ-focused measurements and, where available, serial results. The study does not validate a CRP cut-off, predict subsequent deterioration or bleeding, establish monitoring intervals, or show that adding a test improves outcomes. Neither its odds ratios nor its apparent c-statistics support a new decision rule or a change in treatment.

Several limitations constrain the strength of these conclusions. The retrospective, single-centre cohort contains only 48 patients and reflects a referral population of adults with imported malaria. Selection and referral patterns, the small number of patients with marked renal abnormalities and the long inclusion period limit transportability. The results cannot be assumed to apply to children, endemic populations or other Plasmodium species. Missing procalcitonin, urea and parasitaemia values reduce the secondary-analysis samples and may introduce selection bias. Variation in prior malaria exposure and immune status was not incorporated into the models, limiting interpretation across immune backgrounds. Potential influences such as symptom duration, comorbidities and coinfections were not accounted for comprehensively.

Statistical uncertainty is also substantial. Eight variables were analysed in a small sample, and Bonferroni correction may have obscured weaker associations. Non-significant edges do not establish conditional independence. The approximate t-tests for rank-based partial correlations and the unregularised matrix estimate warrant caution; the bootstrap intervals reported here concern only bivariate correlations, not edge stability. Methods for assessing network accuracy emphasise the importance of this distinction [14]. The secondary analyses were not corrected for multiple testing. Firth penalisation reduces some small-sample estimation problems, but it cannot compensate for the limited information carried by at most 15 severe-thrombocytopenia events; the confidence intervals remain wide and no external validation was performed.

Finally, the study used admission measurements rather than longitudinal trajectories or validated clinical outcome prediction. It cannot establish temporal order, mediation or causal separation of inflammation from organ injury. Larger prospective multicentre studies should examine network uncertainty, incorporate relevant clinical and mechanistic covariates, and test whether any proposed marker combination adds value beyond established severity assessment in an external population. The present results are exploratory observational evidence and are insufficient on their own to support clinical practice recommendations.

Conclusions

In imported falciparum malaria, routine laboratory correlations should be interpreted in the context of systemic inflammation and the organ-related information provided by each test. Partial correlation analysis helps explore which associations persist after adjustment, but it cannot prove whether they are caused by inflammation or are organ-specific. For physicians and laboratory diagnosticians, the practical implication is to assess co-occurring abnormalities alongside clinical findings and serial organ-focused tests, without inferring direct inter-organ injury from correlation alone. This small single-centre study provides hypothesis-generating evidence, not validated diagnostic thresholds, prognostic rules or treatment recommendations. Prospective external validation is required before clinical implementation.

Author contributions

Damian Pikor: Conceptualization, Methodology, Formal analysis, Investigation, Data curation, Visualization, Writing (original draft), Writing (review and editing). Maciej Cymerys: Conceptualization, Supervision, Validation, Writing (review and editing). Małgorzata Paul: Writing (review and editing). All authors read and approved the final manuscript.

Funding

This research received no external funding.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Conflict of interest

The authors declare that they have no conflict of interest.

Ethical approval

The study was conducted in accordance with the Declaration of Helsinki. The Bioethics Committee of Poznań University of Medical Sciences determined that this non-experimental, retrospective study did not require the Committee’s approval under applicable Polish law and the principles of Good Clinical Practice (ruling no. KB-62/26, 28 January 2026).

Consent to participate

Patient consent was waived owing to the retrospective use of fully anonymised routine clinical data.

Consent for publication

Not applicable; the manuscript contains no individually identifiable data.