Neuropsychiatric manifestations of COVID-19 can be clustered in three distinct symptom categories

The study addresses how neuropsychiatric manifestations present and cluster among patients with COVID-19, exploring whether distinct symptom groupings suggest different underlying mechanisms. SARS-CoV-2, initially identified in late 2019, primarily affects respiratory and gastrointestinal systems via ACE2 receptors. β-coronaviruses can involve the central nervous system, potentially via ACE2 expression in the brain and direct entry through the olfactory nerve, consistent with early olfactory symptoms. Prior work often used simple frequency descriptions and reported neurological manifestations in a subset of patients, with rare encephalitis/meningitis. It remains unclear whether olfactory symptoms predict neurological complications or vice versa, and how neuropsychiatric symptoms relate to disease severity and biomarkers. Thus, the authors aimed to systematically evaluate neuropsychiatric symptoms and identify symptom clusters using hierarchical clustering in a large COVID-19 cohort.

Background literature indicates: β-coronaviruses, including SARS-CoV-2, may involve the CNS, possibly via ACE2-mediated mechanisms and olfactory nerve entry. Early olfactory and gustatory dysfunctions are common in mild-to-moderate COVID-19. Prior case series (e.g., Mao et al.) reported neurological manifestations in hospitalized patients (~36%), with rare severe complications like encephalitis and stroke. Reports also describe psychosis, delirium, and neuro-inflammation in COVID-19, and propose mechanisms such as cytokine storm, hypoxia, endothelial/BBB involvement, and retrograde neural transmission. Biomarkers (e.g., WBC, platelets, IL-6) have been investigated for risk stratification. However, few studies used statistical clustering to categorize neuropsychiatric symptoms, motivating the present analysis.

Design and setting: Cross-sectional review of 201 patients with confirmed COVID-19 (52 outpatients, 149 inpatients) at Rasoul Akram Hospital, Tehran, Iran, from March 2019 to April/May 2020. Diagnosis was by positive RT-PCR or typical chest CT (reviewed by three specialists) supported by laboratory tests, per WHO interim guidelines. Ethics approval obtained (IR.IUMS.REC.1399.080); informed consent from patients or relatives. Participants and data collection: Patients with no/mild symptoms were home-quarantined and followed via an online checklist by a trained medical student. Hospitalized patients (moderate to severe symptoms) were examined daily by a neurology resident. Data collected from patients, families, and medical teams included demographics, comorbidities, drug and neuropsychiatric history, COVID-19 symptomatology, and, when indicated, paraclinical data: chest CT, arterial blood gas (ABG), complete blood count (CBC), liver and renal function tests, inflammatory markers, and CPK. Additional tests were ordered based on clinical indications (e.g., O2 saturation <90%, impaired consciousness, hypotension, comorbidity, clinical deterioration, dehydration, intolerance of oral intake). Seizure or stroke during COVID-19 was confirmed by a relevant specialist. Statistical analysis: Conducted in R 3.6.1 (RStudio 1.2.5001). Continuous variables summarized by mean (SD) or range; Wilcoxon rank-sum used for comparisons. For clustering, data were arranged with individuals as rows and symptoms as columns; missing values were removed. Gower dissimilarity (StatMatch package) was computed for categorical attributes. Hierarchical agglomerative clustering used hclust with Ward.D2 linkage (cluster package). A dendrogram depicted cluster structure (Fig. 2), and a scatter plot visualized relationships (factoextra package; Fig. 3). For laboratory predictors, stepwise linear regression (MASS::stepAIC) selected models for WBC, SGPT, SGOT with a two-sided alpha of 0.05 (Table 1). Trends of CBC, LFT/CPK, and ABG were plotted across repeated measurements in subsets of hospitalized patients (n≈20 for CBC/LFT; n≈10 for ABG).

• Symptom frequencies (n=201): muscle weakness 60.2%, dyspnea 56.0%, cough 55.2%, fever 51.2%.
• Neuropsychiatric involvement: 75.1% (151/201) had at least one neuropsychiatric symptom.
• Common neuropsychiatric symptoms: limb force reduction 40.3%, headache 39.8%, anosmia 33.8%, hypogeusia 32.8%.
• Clustering results (six main clusters):
  1. Olfactory: anosmia + hypogeusia.
  2. General constitutional-like neuropsychiatric: dizziness + headache + limb force reduction.
  3. Specific neuropsychiatric/CNS-type: photophobia + mental state change/altered mental status + hallucinations (auditory/visual) + vision problem + speech problem + seizure + stroke + balance disturbance.
  4. Gastrointestinal: diarrhea + nausea (separate from respiratory; closer to neurological, especially olfactory and general neuropsychiatric clusters).
  5. Respiratory: cough + dyspnea.
  6. Constitutional: fever + weakness (closer to respiratory cluster).
• Skeletal muscle manifestations separated into: subjective muscle weakness (clustered with constitutional symptoms) and decreased limb force (clustered with neuropsychiatric symptoms), suggesting potentially distinct mechanisms.
• Laboratory trends: In 20 hospitalized patients, platelet count showed a decrease following an increase over the disease course; CPK surged at the third testing; in 10 patients ABG factors were trended (details shown in Fig. 4).
• Regression analyses (subset with labs): • WBC predictors: Model R^2=0.79, F(19,13)=2.69, p=0.03. Significant predictors included onset (β=-0.20, p=0.02), dyspnea (β=3.96, p=0.02), cough (β=-3.66, p=0.02), nausea (β=-4.12, p=0.01), diarrhea (β=5.77, p=0.006), balance disturbance (β=-8.83, p=0.001), limb force reduction (β=-4.01, p=0.05), anosmia (β=4.43, p=0.02), hypogeusia (β=-7.28, p=0.01), disorientation (β=14.75, p<0.001); auditory hallucination showed a large positive estimate (β≈13.27) with reported p=0.003. • SGPT predictors: R^2=0.59, F(10,27)=2.69, p=0.02. Significant variables: fever (β=46.18, p=0.01), limb force reduction (β=-53.49, p=0.01), auditory hallucination (β=178.23, p=0.002), visual hallucination (β=-154.01, p=0.01), anosmia (β=-105.16, p=0.002), hypogeusia (β=110.00, p=0.004). Trends for cough and balance disturbance approached significance. • SGOT predictors: R^2=0.17, F(1,36)=2.69, p=0.008; photophobia was the sole significant predictor (β=171.89, p=0.008).
• Rare severe CNS events: seizure 2%, stroke 1%. Hallucinations in ~11%; photophobia ~9%.

The study demonstrates that neuropsychiatric symptoms in COVID-19 are common and cluster into three distinct categories, suggesting heterogeneous pathophysiological mechanisms. The separation of olfactory symptoms (anosmia, hypogeusia) from other neuropsychiatric manifestations supports the hypothesis of a distinct mechanism, potentially via retrograde transmission through the olfactory bulb, rather than systemic processes. The clustering of serious CNS-type manifestations (e.g., photophobia, altered mental status, hallucinations, seizures, stroke) into a single group may reflect a specific CNS-involving subtype of disease and warrants heightened clinical attention. Gastrointestinal symptoms formed a cluster distinct from respiratory symptoms and were closer to neurological clusters, aligning with shared ACE2-related pathways and systemic inflammatory effects. Skeletal muscle symptoms partitioned into constitutional weakness versus neurologic limb force reduction, indicating potentially different neuromuscular etiologies. Associations between selected symptoms and laboratory markers (WBC, SGPT, SGOT) further support links between clinical phenotypes and systemic inflammation or organ involvement. Compared with prior reports (e.g., Mao et al.), the higher prevalence of neuropsychiatric symptoms may relate to broader symptom capture (e.g., hallucinations, photophobia, speech problems) and differing definitions of muscle involvement. Clinically, the findings argue for comprehensive neuropsychiatric assessment across all three categories, prioritizing urgent evaluation for specific CNS-type symptoms, and tailoring management strategies accordingly.

Neuropsychiatric presentations in COVID-19 are common (75% in this cohort) and heterogeneous, clustering into three categories: olfactory symptoms; general constitutional-like neuropsychiatric symptoms; and specific CNS-type neuropsychiatric symptoms. Olfactory dysfunction appears distinct from other neuropsychiatric clusters, implying different mechanisms of CNS involvement. Skeletal muscle manifestations may reflect either constitutional or neuropsychiatric processes. These insights inform clinical assessment and triage, suggesting that specific CNS-type symptoms should prompt urgent investigation (e.g., neuroimaging), while general constitutional-like symptoms may require supportive care. Future research should use larger cohorts with comprehensive laboratory panels, neuroimaging, viral genotyping, and HLA mapping to clarify mechanisms, predictors of severe CNS involvement, and targeted therapies.

• Cross-sectional design limits causal inference and temporal relationships.
• Some neuropsychiatric symptoms may stem from medications, electrolyte disturbances, hypoxia, or comorbidities rather than direct viral effects.
• Laboratory tests were ordered based on clinical indication; not all inpatients had complete paraclinical data, reducing power and introducing selection bias.
• Small subsamples for lab trend and regression analyses (≈20 for CBC/LFT; ≈10 for ABG) limit generalizability.
• Brain imaging was reserved for critical cases; some strokes may have been underdiagnosed.
• Lack of viral genotyping and patient HLA mapping limits mechanistic insights.
• The additional affiliation indicator (e.g., superscript 8) was not detailed in the provided affiliations list.