Clinical scoring systems for diagnosing tuberculous meningitis: A systematic review

Abstract

BACKGROUND

Tuberculous meningitis (TBM) is a severe extrapulmonary complication of tuberculosis. Early diagnosis is crucial in improving patient outcomes, albeit difficult given the non-specific clinical presentation of early stage TBM. Numerical diagnostic scoring systems may be useful diagnostic tools for clinicians, particularly in resource-constrained settings.

AIM

To describe existing TBM diagnostic scores with the primary focus being on scores with external validation.

METHODS

PubMed and EMBASE were searched from inception to 21 March 2024. Cross-sectional studies with numerical TBM scoring systems were included. Literature pertaining to non-human, non-live subjects, single diagnostic tests, papers with no gold standard diagnosis, and non-English papers were excluded.

RESULTS

We reviewed 21 unique scoring systems, of which 4 have been externally validated. Key features of the scores (clinical, laboratory, microbiological, histopathological and radiological criteria, interpretation), outcomes (sensitivity, specificity) and study population (location, demographics, subgroups) were described.

CONCLUSION

We suggest Thwaites (2002) and Marais (2010) as first-line TBM screening tests due to high sensitivities and extensive external validation, however clinicians should consider scores specific to their patient population.

Key Words: Tuberculous meningitis; Central nervous system tuberculosis; Diagnostic scores; Diagnostic models; Sensitivity and specificity

Core Tip: Tuberculous meningitis (TBM) is a severe complication of tuberculosis. Early diagnosis is crucial in improving patient outcomes, albeit difficult given the non-specific signs and symptoms of early TBM. Numerical diagnostic scoring systems may be a useful early diagnostic tool for clinicians, particularly in resource-constrained settings. Of all the scores currently published, we suggest using Thwaites (2002) and Marais (2010) as first line TBM screening tests due to their high sensitivities and extensive external validation. Nevertheless, clinicians are encouraged to select scores in accordance with the patient characteristics of the population of interest.

INTRODUCTION

Tuberculosis (TB) is the second leading cause of death from a single infectious agent, as reported in 2023[1]. Tuberculous meningitis (TBM) is the most severe form of central nervous system TB, with high morbidity and risk of neurodevelopmental sequelae. Even amongst patients receiving appropriate treatment, the mortality of the disease remains high. A 2021 modelling study investigating TBM globally found that TBM case fatality among patients undergoing treatment in 2019 was on average 27%[2]. An estimated 48% of incident TBM cases in 2019 resulted in death[2]. Timely diagnosis and treatment is key to improving patient outcomes[3]. Unfortunately, the diagnosis of TBM remains difficult. Early prodromal stage TBM presents with non-specific symptoms that frequently overlap with non-TBM (including bacterial, viral, fungal and aseptic causes)[4]. Furthermore, the time course of disease is variable, ranging from several days to six months[5].

World Health Organization-recommended gold standard diagnostic tests using cerebrospinal fluid (CSF) include: (1) Acid-fast bacilli (AFB) smear; (2) M.TB cultures; and (3) Nucleic acid amplification tests (NAATs), such as Xpert MTB/RIF Ultra. However, these tests can also contribute to diagnostic challenges. Sensitivities of AFB smears and cultures are suboptimal. Conventional cultures can take up to months in certain areas. NAATs, while more efficient, have higher processing costs and limited availability in resource-constrained areas[6].

Given these challenges, various numerical diagnostic scores have been produced with the aim of aiding diagnosis through a combination of clinical, laboratory, microbiological, histopathological and radiological criteria. Some notable scoring systems include that of the Lancet/Marais score (area under the receiver operating characteristic curve 0.90 to 0.979) and Thwaites (area under the receiver operating characteristic curve 0.666 to 0.99) In this study, we aim to describe key features of available TBM diagnostic scores, focusing on externally validated scores. Features described include accuracy (sensitivity, specificity), strengths and limitations.

MATERIALS AND METHODS

Study questions

This study seeks to serve as an overview of the scoring systems that are currently available for TBM diagnosis. In comparing the various systems, we seek to answer the following questions: (1) How do the scores vary in terms of sensitivity, specificity and applicability to various populations? (2) What are the strengths and limitations of existing scores affecting their utility in the real world? and (3) What is the role of diagnostic scoring systems in today’s setting and the future?

Search strategy

The literature search was conducted in accordance with the PRISMA guidelines. We searched PubMed and EMBASE from conception to 21 March 2024, using combinations of: “tuberculosis meningitis”, “central nervous system tuberculosis”, “cerebral tuberculosis”, “brain tuberculosis”, “diagnosis”, “score”, “model”, “criteria”, “rule” and “test”. The protocol for this study was registered in PROSPERO (CRD42024518114).

Study selection

Articles were screened independently by three researchers (Teo E, Kung K, and Chen S) first by title and abstract, then by full-text using eligibility criteria described below. The references of included papers were screened for relevant articles. Disagreement was resolved through discussion, reanalysis of data and consensus. A systematic review of literature was conducted for all existing cross sectional studies containing data on TBM scoring systems.

Exclusion criteria

Exclusion criteria was: (1) Non-human or non-live subjects; (2) Patients not diagnosed with TBM; (3) Papers not related to TBM diagnostic tests; (4) Papers with only single diagnostic tests; (5) Papers without a numerical scoring system; (6) Papers without a gold standard of TBM diagnosis (positive CSF culture/polymerase chain reaction/NAATs); (7) Articles not in English; (8) Papers with chronic meningitis as the main patient population; (9) Systematic reviews, review articles, editorials, letters, single case studies, studies with no specificity; and (10) Papers where the full text was unavailable. 4 studies were excluded as the full text was unavailable. Figure 1 represents a summary of the study selection process.

Figure 1 PRISMA flow chart of the study selection process. TB: Tuberculosis; TBM: Tuberculous meningitis.

Quality assessment

The Joanna Briggs Institute Critical Appraisal tool (Supplementary Table 1) was used to assess studies’ methodological quality and determine the extent of possible bias in terms of design, conduct and analysis. The tools were applied by three researchers independently, with disagreements resolved through discussion and consensus.

Key definitions

Numerical scoring systems refer to a set of criteria with a numerical value per component totaling to a final score. Modifications to scoring systems include alterations to cut-off values of individual components which originally involve a range of values. Papers which added extra components with their own numerical values to existing scores were assigned their own score. External validation is defined as the process of validating results using a separate group of patients from a different state or country from the original derivation cohort.

RESULTS

30 publications describing 21 scoring systems and 7127 patients were included for final analysis.

Scores with external validation

We reviewed 4 externally validated scoring systems: Thwaites (2002), Hristea (2012), Marais (2010); and Kumar (1999), with the year of development in parentheses. Throughout literature, Thwaites (2002) is also referenced as the Vietnam rule and Marais (2010) as the Lancet scoring classification; the former names will be used in this review for consistency. Details of each score and their interpretation are in Table 1. Notably, only Marais (2010) included radiological and microbiological criteria. Other scores only required simple laboratory investigations (CSF and serum analysis). Cohort populations also varied in age - Kumar (1999) used a paediatric population, Hristea (2012) a mixed paediatric and adult population and Thwaites (2002) an adult population.

Table 1 Details of scores with external validation.

Ref.	Country	Internal validation	Clinical criteria	Laboratory criteria	Microbiological/histopathological criteria	Radiological criteria	Interpretation
Thwaites et al[31], 2002	Vietnam	AUC of ROC: 0.990	Age: (1) ≥ 36 years: 2 points; and (2) < 36 years: 0 points	CSF analysis (103/mL)	None	None	TDI: DI (age) + DI (blood white-cell count) + DI (history of illness) + DI (cerebrospinal fluid white-cell count) + DI (cerebrospinal fluid% neutrophils)
		Sensitivity 97.0%; specificity: 91.0%		CSF total white cell count
				≥ 900: 3 points; < 900: 0 points			TBM: TDI ≤ 4
				CSF% neutrophils: ≥ 75: 4 points; < 75: 0 points
		After internal validation with test data population (n = 75)	History of illness: (1) ≥ 6 days: -5 points; and (2) < 6 days: 0 points	Blood white cell count (103/mL): (1) ≥ 15000: 4 points; and (2) < 15000: 0 points			BM: TDI > 4
		Sensitivity: 86.0%; specificity: 79.0%
Marais et al[32], 2010	NIL	NIL	Entry criteria: Symptoms and signs of meningitis including at least one: Headache, irritability, vomiting, fever, neck stiffness, convulsions, focal neurological deficits, altered consciousness, or lethargy	CSF analysis 1 point each, maximum score: 4 points	Extraneural TB: Maximum score: 4 points. Imaging suggestive of extraneural TB (see next column). AFB identified or M.TB cultured from another source (e.g., sputum, lymph node, gastric washing, urine, blood culture): 4 points. Positive commercial M.TB NAAT from extra-neural specimen: 4 points	Cerebral imaging: Maximum score: 6 points: (1) Hydrocephalus: 1 point; (2) Basal meningeal enhancement: 2 points; (3) Tuberculoma: 2 points; (4) Infarct: 1 point; (5) Pre-contrast basal hyperdensity: 2 points; and (6) Imaging suggestive of extraneural TB: Chest radiograph suggestive of active TB: Signs of TB: 2 points. Miliary TB: 4 points. CT/MRI/ultrasound evidence for TB outside CNS: 2 points	Definite TBM
							Clinical entry criteria
							At least 1: (1) AFB seen in CSF; (2) M.TB cultured from CSF; and (3) CSF positive commercial NAAT
			Clinical criteria: Maximum score: 6 points	Clear appearance			“AND/OR” AFB seen in context of histological changes consistent with TB in the brain or spinal cord with suggestive symptoms or signs and CSF changes, or visible meningitis (on autopsy)
			Symptom duration > 5 days: 4 points	Cells: 10-500 per μL			Probable TBM
							Clinical entry criteria: (1) Total diagnostic score (with ≥ 2 point from laboratory or radiological criteria); (2) Score ≥ 10 (when cerebral imaging unavailable); and (3) Score ≥ 12 (when cerebral imaging available)
			Systemic symptoms suggestive of TB (at least 1): 3 points: (1) Weight loss or poor weight gain in children; (2) Night sweats; and (3) Persistent cough > 2 weeks	Lymphocytic predominance (> 50%)			Exclusion of alternative diagnoses
			History of recent (within past year) close contact with an individual with pulmonary TB or positive TST or IGRA (only in children < 10 years old): 2 points	Protein concentration > 1 g/L			Probable TBM
							Clinical entry criteria: (1) Total diagnostic score (with ≥ 2 points from laboratory or radiological criteria); (2) Score ≥ 10 (when cerebral imaging unavailable); and (3) Score ≥ 12 (when cerebral imaging available)
			Focal neurological deficit (excluding cranial nerve palsies): 1 point	CSF: Plasma glucose ratio < 50% or an absolute CSF glucose concentration < 2.2 mmol/L			Exclusion of alternative diagnoses
			Cranial nerve palsy: 1 point				Possible TBM: (1) Clinical entry criteria; (2) Total diagnostic score: Score: 6-9 (when cerebral imaging is not available); score: 6-11 (when cerebral imaging is available); and (3) Exclusion of alternative diagnoses. Possible TB cannot be diagnosed or excluded without doing a lumbar puncture or cerebral imaging
			Altered consciousness: 1 point				Non-TBM: Alternative diagnosis established, without a definitive diagnosis of TBM or other convincing signs of dual disease
Hristea et al[33], 2012	Romania	AUC of ROC: 0.977	DSBA ≥ 5 days: Fever; headache; vomiting; neck stiffness	CSF analysis	None	None	For patients with clear CSF meningitis, use this equation to calculate the probability of TBM: Ln (ODDS) = 3.615 × DSBA + 2.680 × (neurological stage) + 3.414 × (glucose ratio) + 1.669 × (CSF protein) - 7.612
		Sensitivity: 92.0%; specificity: 94.0%	Neurological stage II and III (defined based on Medical Research Council Definitions)	CSF/blood glucose ratio < 0.5
			Stage II: Lethargy, confusion, mild focal signs (cranial nerve palsy, hemiparesis)	CSF protein level > 100 mg/dL			Scoring system: (1) DBSA ≥ 5 days: 3 points; (2) Neurological stages II/III (instead of I): 2 points; (3) Glucose ratio < 0.5: 3 points; and (4) CSF protein > 100 mg/dL: 1 point
		PPV: TBM: 89.1%; VM: 95.6%	Stage III: Delirium, stupor, coma, seizures, multiple cranial nerve palsies, dense hemiplegia or paraparesis
Kumar et al[34], 19991	India	One or more predictor variables present: Total n = 100; TBM n = 65; non-TBM n = 35; sensitivity: 98.4%; specificity: 43.5%	Prodromal stage ≥ 7 days	CSF analysis: CSF leukocytes < 50% polymorphs	None	None	TBM ruled in with high degree of certainty: ≥ 3 of 5 criteria present
		Two or more predictor variables present: Total n = 59; TBM n = 51; non-TBM n = 8; sensitivity: 77.2%; specificity: 87.0%	Fundal optic atrophy				Highly unlikely TBM: None of the 5 criteria present
		Three or more predictor variables present: Total n = 37; TBM n = 36; non-TBM n = 1; sensitivity: 54.5%; specificity: 98.3%	Focal neurological deficits				Use scoring system and ROC plot if ≥ 2 out of 5 criteria present (best cut off: 34): (1) Prodromal stage ≥ 7 days: 23; (2) Fundal optic atrophy: 2; (3) Focal deficits: 18; (4) EP movements: 29; and (5) CSF leucocytes < 50% polymorphs: 34
			Extrapyramidal movements

¹Formulated with paediatric population.

AUC: Area under the receiver operating characteristic curve; ROC: Receiver operating characteristic; CSF: Cerebrospinal fluid; TDI: Total diagnostic index; DI: Diagnostic index; TBM: Tuberculous meningitis; BM: Bacterial meningitis; NIL: Zero; TST: Tuberculosis skin test; IGRA: Interferon gamma release assay test; TB: Tuberculosis; NAAT: Nucleic acid amplification test; CT: Computed tomography; MRI: Magnetic resonance imaging; CNS: Central nervous system; AFB: Acid fast bacilli; DSBA: Duration of symptoms before admission; non-TBM: Non-tuberculous meningitis; EP: Epileptic.

Thwaites (2002), Hristea (2012), and Kumar (1999) were internally validated with their respective test data populations. Internal validation refers to using patients from the developmental population to investigate the reproducibility of the score[7]. Of the 3 internally validated scoring systems, Kumar (1999) had the highest sensitivity of 98.4%.

An overview of the external validations of the 4 scores were detailed in Table 2. External validation looks into the generalisability of scores to various populations[7]. Each score provided a different interpretation of final numerical values. For instance, Marais distinguished between definite, probable and possible TBM. Sensitivity and specificity in Table 2 were calculated based on the score that reflected the lowest threshold for treatment, i.e., the score at which clinicians would be likely to initiate anti-TBM treatment.

Table 2 Overview of scores with external validation.

Ref.	Sensitivity (%)	Specificity (%)	Country of external validation population	Clinical criteria	Laboratory criteria	Microbiological/histopathological criteria	Radiological criteria
Thwaites et al[31], 2002	Cut off of ≤ 3.5: TBM vs non-TBM: 78.0	Cut off of ≤ 3.5: TBM vs non-TBM: 63.0	Malawi	√	√	X	X
	TBM vs PM: 95.0	TBM vs PM: 23.5	Pakistan
	Cut off of ≤ 4: TBM vs non-TBM: 78.0-100	Cut off of ≤ 4: TBM vs non-TBM: 40.8-43.0	Malawi, Turkey, Qatar
	TBM vs BME: 95.8	TBM vs BME: 0.680	Turkey
	TBM vs PM: 96.0	-	India
	TBM vs BM: 95.8-98.2	TBM vs BM: 43.6-87.9	China, Turkey, India
Marais et al[32], 2010	Cut off of ≥ 6: TBM vs non-TBM: 66.7-100	Cut off of ≥ 6: TBM vs non-TBM: 48.7-85.5	Brazil, Turkey, China, Qatar	√	√	√	√
	TBM vs BM: 93.0-97.0	TBM vs BM: 56.0-68.0	China
	TBM vs BME: 96.8	TBM vs BME: 3.06	Turkey
	Cut off of ≥ 8: TBM vs non-TBM: 84.9	Cut off of ≥ 8: TBM vs non-TBM: 77.2	Brazil
	Cut off of ≥ 10: TBM vs non-TBM: 100	Cut off of ≥ 10: TBM vs non-TBM: 40.8	Brazil
Hristea et al[33], 20121	Cut off of ≥ 6: TBM vs non-TBM: 100	Cut off of ≥ 6	Turkey	√	√	X	X
	TBM vs VM: 96.7	TBM vs VM: 81.1	Romania
	TBM vs CM: 86.6	TBM vs CM: 27.7	Romania
	Cut off of ≥ 9: TBM vs VM: 51.6	Cut off of ≥ 9: TBM vs VM: 100	Romania
Kumar et al[34], 19991	Cut off of ≥ 3: TBM vs non-TBM: 37.0	Cut off of ≥ 3: TBM vs non-TBM: 58.1	India	√	√	X	X

¹Validated in the paediatric population.

TBM: Tuberculous meningitis; non-TBM: Non-tuberculous meningitis; BME: Bacterial meningoencephalitis; BM: Bacterial meningitis; VM: Viral meningitis; CM: Cryptococcal meningitis; PM: Pyogenic meningitis.

Generally, Thwaites (2002) was most sensitive (78.0%-100.0%), while Kumar (1999) was the least (37.0%). Some studies modified the existing score to better suit local populations. Thwaites (2002) was frequently modified in externally validated cohorts. An adjustment in cut-off from ≤ 4 to ≤ 3.5, where patients who scored ≤ 3.5 points were more likely to have TBM, was noted in 2 papers with patients from Malawi and Pakistan[8,9]. The cut-off of each perimeter of Thwaites (2002) was also changed in an Indian population in a study by Vibha et al[10], seen in Supplementary Table 2.

Scores without external validation

The remaining 17 scores were not externally validated as of time of writing. Table 3 provides an overview of the scores from highest to lowest sensitivity. Of the 17, Yang (2019) had the highest sensitivity (97.0%-98.0%). Further details of each score are in Supplementary Tables 3-24. Of note, Solomons created 2 different scoring systems, Solomons A (2016) and Solomons B (2016), the latter being a simplified 3-variable score. Sensitivity and specificity in Table 3. Lu (2021), Luo (2021) and Yang (2020) had additional patient cohorts internally validating their scores. These additional cohorts comprised patients in the same province as the original cohort. The accuracy of these additional cohorts are provided in Supplementary Tables 13, 14, and 23.

Table 3 Overview of scores without external validation.

Ref.	Sensitivity	Specificity	Clinical criteria	Laboratory criteria	Microbiological/histopathological criteria	Radiological criteria	Country
Yang et al[35], 2019	97.0-98.0	81.0-82.0	√	√	X	X	China
Goenka et al[36], 2018	93.0-100	89.0-93.0	√	√	√	X	South Africa
Wen et al[37], 2022	96.0	86.0	√	√	X	X	China
Vibha et al[10], 2012	95.7	97.6	√	√	X	X	India
Ersoy et al[38], 2012	95.5	100	√	√	X	X	Turkey
Solari et al[39], 2018	94.9	52.6-94.7	√	√	X	X	Peru
Chusri et al[40], 2018	92.0	89.0	√	√	X	√	Thailand
Lee et al[41], 2018	89.4	80.4	√	√	X	X	South Korea
Solomons et al[42] 20161	97.0	48.0-78.0	√	√	√	√	South Africa
	79.0	48.0-78.0	√	√	√	√
He et al[43], 2021	87.0	94.0	√	√	X	X	China
Dendane et al[44], 2013	86.0-88.0	93.0-95.0	√	√	√	√	Morocco
Lu et al[45], 2021	81.0-85.8	86.3-87.7	√	√	X	√	China
Luo et al[46], 2021	81.6	91.8	X	√	X	X	China
Fong et al[24], 2023	81.5	88.1	√	√	X	√	China
Handryastuti et al[47], 20231	47.1	95.1	√	X	X	√	Indonesia
Patel et al[48], 2010	38.0	93.0	√	√	X	X	South Africa

¹Validated in the paediatric population. Refer to Supplementary Table 3 for further details.

DISCUSSION

Strengths of diagnostic scores

We reviewed 21 existing numerical TBM diagnostic scoring systems. Encouragingly, the externally validated scores display high sensitivity. Across the board, Thwaites (2002) was the most sensitive, achieving 86.0%-97.0% sensitivity with internal validation and 78.0%-100% sensitivity with external validation in 9 different study settings in 6 countries. Sensitivity of Marais (2010) was also relatively high, ranging from 66.7%-100% in 6 study settings in 4 countries. These high sensitivities render them suitable screening tests that can potentially rule out TBM when used in conjunction with clinical acumen, which is particularly important in a high mortality and morbidity disease[11].

TB incidence rises as gross domestic product per capita falls and rates of poverty increase[12]. In such settings, scores are especially useful in establishing a quick diagnosis for early initiation of empirical treatment - the most significant factor in reducing mortality[13]. A significant advantage of 3 of 4 externally validated scores -Thwaites (2002), Hristea (2012) and Kumar (1999) is that they use a combination of clinical criteria and readily-available biochemical tests. Marais (2010) accounts for its use of less available microbiological and radiological criteria by adjusting score cut-off values for when cerebral imaging is unavailable, therefore keeping it accessible for both resource-constrained and non-resource-constrained settings. Microbiological findings are also not essential to diagnose possible or probable TBM. Even in the case where such investigations are unavailable, a likely TBM diagnosis can still be made, and empirical treatment can be started. Furthermore, the scores do not require complex digital programmes for application or interpretation. They can thus be integrated into clinical practice seamlessly without skill training or technology upgrades.

Limitations of diagnostic scores

Firstly, scoring systems cannot replace definitive tests, and should act as a guiding tool instead. Some patients will inevitably be classified as having intermediate risk of TBM or have an indeterminate score[6]. In such cases, diagnostic scores may ultimately be of little utility as further evaluation and/or testing by clinicians will still be necessary. Additionally, score models only take into consideration the sign or symptom at the time of presentation. Most of the scores do not include a follow-up component, although clinical, laboratory, microbiological and radiological findings may vary over the disease course and alter scores, particularly in ambiguous presentation[3]. Thus, in ambiguous presentations, we propose clinicians re-calculate patients’ numerical scores throughout admission for disease progression or atypical findings pointing to other diagnoses.

Next, not all scores accounted for significant TBM determinants in formulation and validation. Known risk factors associated with TB include undernourishment, human immunodeficiency virus (HIV) co-infection, alcohol use, smoking and diabetes[1,12]. Of these, HIV co-infection is the most significant risk factor - HIV-positive patients are 15 times to 21 times more likely to develop TB disease[14]. However, only Goenka (2018), which has yet to be externally validated, included HIV co-infection as a criterion. Several studies listed HIV co-infection as an exclusion criteria, limiting their utility in HIV-positive patients. One study in Malawi investigating Thwaites concluded that the score was not clinically appropriate in settings with high HIV-prevalence[8]. The score performed sub-optimally (78% sensitive, 43% positive) in their cohort, with all 12 patients with cryptococcal meningitis giving false positives, a finding attributed to the high prevalence of HIV and HIV-associated cryptococcal meningitis in Malawi. Age is another significant factor in TBM. Young children infected with TB are at high risk of disseminated TB and TBM incidence peaks in children between the ages of 2 years and 4 years[15]. Scores formulated with adult populations may perform differently and have yet to be validated with paediatric populations and vice versa. Of the externally validated scores, only Hristea (2012) was formulated with a population of both adults and children. Therefore, further research on the use of scores in these vulnerable populations should be done. Clinicians operating in settings with high HIV prevalence or treating paediatric patients may consider scores that have been validated in similar populations.

Additionally, scores may be limited by the heterogeneity of the papers and their respective study locations and populations. TB prevalence and strains vary greatly from country to country (and within countries, from region to region). Certain strains of M.TB have been reported to be more likely than others to progress to TBM[16]. Similarly, the prevalence of non-TBM aetiologies varies in different regions. Cryptococcus, which causes cryptococcal meningitis is endemic in tropical, subtropical and temperate climatic regions[17]. Meanwhile, streptococcus pneumoniae is the most common bacterial cause of meningitis in the African and East Mediterranean region as compared to Neisseria meningitidis in the European region[18]. Local prevalence of TBM and non-TBM causes should also be considered when applying these scores.

Finally, 18 of 30 included studies were carried out retrospectively using medical records. Data collection was not standardized in several retrospective studies when investigations were not carried out routinely in all patients. Furthermore, potential inaccuracies during the verification of clinical characteristics, such as signs and symptoms elicited on examination, could have led to misclassification bias. That being said, the remaining 12 studies were either prospective studies or a combination of both. These studies investigating the same scores had comparable results. As such, the effect of retrospective analysis on the accuracy of results is likely minimal for externally validated scores. However, 8 of the scores without external validation [Chusri (2018); Ersoy (2012); Fong (2023); Goenka (2018); Handryastuti (2023); Solomons (A) (2016); Solomons (B) (2016); Wen (2022)] were formulated retrospectively and have yet to be validated or compared with prospective data.

Future directions and research

Further validation with different study populations should be done to improve credibility. Existing scoring systems may be mostly applicable to the paper’s study populations with specific demographics and not generalisable to other populations. This is particularly relevant in the case of the 17 scores lacking external validation reviewed in this study. As aforementioned, an immunocompromised state alters the differential diagnoses in a patient with meningitis, which is not accounted for in some of the existing scoring systems[19]. Validation of scores with HIV-positive patients and other immunocompromised cohorts, such as organ transplant patients, is another possibility.

Incorporating CSF and serum biomarkers in scoring systems is an avenue for further research[20]. For example, adenosine deaminase (ADA) was found to be elevated in the CSF of both adult and paediatric TBM patients[21]. 2 scores without external validation, Chusri (2018) and Solari (2018), use CSF ADA as a criterion. A study by Bhatnagar et al[22] also reported that including positive CSF ADA improved the sensitivity and specificity of Kumar (1999) for all TBM stages. Another promising example is tuberculostearic acid, which can promptly detect the presence of M.TB with high sensitivity and specificity in one to two days[23]. When incorporated into a diagnostic model, diagnostic accuracy was found to be comparable to Marais (2010)[24]. As of the time of writing, tuberculostearic acid is only incorporated in Fong (2023). Serum tests can also support TBM diagnosis. One 2019 study found that the combination of 3 serum biomarkers (adipsin, Aβ42, and interleukin-10) was 82.6% sensitive in diagnosing TBM in children[25]. A similar 2016 study found that a 7-marker host serum protein biosignature (C-reactive protein, transthyretin, interferon-γ, complement factor H, apolipoprotein-A1, inducible protein 10 and serum amyloid A) was 93.8% sensitive in diagnosing TB in adults[26].

The role of other supporting tests can also be explored further. Tests such as tuberculin skin test and interferon-gamma release assays support TB diagnosis, although their role in definitive diagnosis is controversial[27-29]. Due to less cross-reactivity with prior bacillus Calmette-Guérin vaccination, interferon-gamma release assays are more specific than tuberculin skin test and can thus serve as a possible criterion in diagnostic scores[30].

CONCLUSION

Numerical diagnostic scoring systems are easily accessible and useful adjuncts to definitive TBM tests. We suggest using Thwaites (2002) and Marais (2010) as screening tests, as they have been externally validated in multiple settings and show the highest sensitivities. Nevertheless, there is much room for improvement in terms of additional external validation in other specific patient cohorts, especially HIV-positive and paediatric populations. Clinicians should exercise their best judgement in selecting an appropriate scoring system based on the characteristics of the population of interest.