Research published between 21 July and 3 August 2026

This fortnight’s fungal diagnostics research highlights two complementary challenges: making established diagnostic tests accessible and developing better ways to recognise high-risk patients before definitive results are available.

A major multinational survey found substantial inequalities in access to fungal diagnostics across Latin America and the Caribbean. Other studies evaluated prediction models for talaromycosis and invasive pulmonary aspergillosis (IPA), while a smaller study proposed a diagnostic and severity-assessment workflow for non-HIV Pneumocystis pneumonia.

Major inequalities in access to fungal diagnostics

A survey published in Nature Communications examined fungal diagnostic capacity, antifungal availability and therapeutic drug monitoring across Latin America and the Caribbean.

The researchers received responses from 619 institutions in 23 countries. Although conventional fungal culture was available in 90% of participating centres, access to more advanced diagnostic methods was much more limited:

• Galactomannan testing was available in 41% of centres.
• β-D-glucan testing was available in 29%.
• Molecular fungal testing was available in 23%.
• Therapeutic drug monitoring was available in 36%.

Diagnostic and treatment provision was significantly better in countries with a gross domestic product per person above US$10,000, in transplant centres and in institutions caring for people living with HIV.

The survey also found important limitations in access to tests for endemic fungal infections, including histoplasmosis, coccidioidomycosis and paracoccidioidomycosis. These infections may therefore remain undiagnosed or be recognised only after significant delays.

Access to antifungal treatment was similarly uneven. Liposomal amphotericin B was available in only 38% of centres, voriconazole in 57% and posaconazole in 33%.

Why is this important?

Developing new fungal diagnostics is only part of the solution. Many centres still lack consistent access to tests that are already recommended and widely used elsewhere.

Fungal culture remains valuable, but it can be slow and may have limited sensitivity. Restricted access to antigen, biomarker and molecular tests can delay diagnosis, particularly when patients are severely immunocompromised or have already received antifungal treatment.

The study provides evidence that national income, laboratory infrastructure and institutional complexity continue to influence whether patients can receive timely fungal diagnosis and appropriate treatment.

The survey was voluntary and relied on self-reported information. Participation was also uneven between countries, with approximately one-third of responses coming from Colombia. Nevertheless, its large scale makes it an important assessment of regional diagnostic capacity.

Read the multinational survey in Nature Communications

Machine learning may help recognise talaromycosis without skin lesions

Talaromycosis is a serious systemic fungal infection caused by Talaromyces marneffei. It occurs mainly in Southeast Asia and southern China and particularly affects people with advanced HIV.

Characteristic skin lesions can provide an important diagnostic clue, but some patients do not develop them. In these cases, talaromycosis may resemble tuberculosis, Pneumocystis pneumonia, cryptococcosis or other opportunistic infections.

Researchers developed a machine-learning model using records from 1,009 people with HIV who were admitted to hospitals in China with suspected opportunistic infections. All had CD4 counts below 200 cells/μL and no characteristic talaromycosis skin lesions.

The definitive diagnosis was based on isolation of T. marneffei from clinical samples or identification in biopsy tissue. Twelve routinely available variables were incorporated into the final model, including:

• albumin;
• haemoglobin;
• white blood cell and lymphocyte counts;
• platelet count;
• CD4 T-cell count;
• C-reactive protein;
• liver enzyme measurements;
• age; and
• peripheral or abdominal lymphadenopathy.

Five machine-learning approaches were compared. A support-vector-machine model performed best, producing an area under the receiver operating characteristic curve—or AUC—of 0.809 in model development. An AUC of 0.5 represents performance no better than chance, while 1.0 represents perfect discrimination.

The model was then tested using 305 patients from an independent hospital. In this external-validation group, the AUC was 0.921, overall accuracy was 85.3% and the F1 score was 0.819.

Could it be used clinically?

Independent validation is an important strength and is often missing from medical machine-learning studies. The researchers also used interpretation methods to show how individual clinical variables contributed to each prediction and made the model available as an online tool.

However, this remains a retrospective study using Chinese hospital records collected between 2010 and 2019. The model needs prospective evaluation and testing in other endemic countries, healthcare systems and patient populations.

The unexpectedly higher performance in the external cohort should also be investigated. It may reflect genuine robustness, but it could result from differences in disease prevalence, patient selection or case complexity.

The model should therefore be regarded as a method of identifying patients who require urgent fungal investigation. It does not replace fungal culture, histopathology or validated antigen and molecular tests.

Read the talaromycosis study in Mycopathologia

A nine-factor model for estimating IPA risk

A separate study developed a clinical model for estimating the likelihood of invasive pulmonary aspergillosis among patients with pulmonary infections.

The nested case-control study was drawn from a ten-year cohort of 27,100 patients. The model-development dataset included 1,002 people with proven or probable IPA and 2,004 randomly selected pneumonia controls.

Nine independently associated factors were incorporated into the model:

• bronchiectasis;
• previous pulmonary tuberculosis;
• diabetes;
• positive serum galactomannan;
• mechanical ventilation;
• connective-tissue disease;
• positive serum β-D-glucan;
• sputum findings; and
• neutrophil-to-lymphocyte ratio.

Bronchiectasis produced the strongest association with IPA, with an odds ratio of 7.07. Previous tuberculosis and diabetes were each associated with approximately twice the odds of IPA.

The model achieved an AUC of 0.73 in the training data and 0.75 in the internal testing group. This represents moderate rather than high discriminatory performance.

Important limitations

The model was tested using a held-back portion of the same underlying patient population, rather than in an independent hospital or country. External validation is therefore required before it could be recommended for clinical use.

There is also potential circularity because galactomannan contributes to accepted definitions of probable IPA and was then included as one of the predictors of that diagnosis. The case-control design may make the predicted absolute risks difficult to transfer directly to an ordinary hospital population.

The nomogram may eventually help clinicians recognise combinations of risk factors that justify earlier fungal investigation. At present, however, it should not be interpreted as a standalone diagnostic test or used to begin treatment without supporting clinical, radiological and microbiological evidence.

Read the IPA prediction study in Medical Mycology

Proposed workflow for non-HIV Pneumocystis pneumonia

Non-HIV Pneumocystis jirovecii pneumonia can progress rapidly in people receiving corticosteroids, chemotherapy, transplantation treatment or other forms of immunosuppression.

A single-centre observational study of 49 adults with clinically confirmed Pneumocystis pneumonia proposed a workflow combining:

• recognition of a high-risk host;
• compatible symptoms and CT appearances;
• molecular testing of bronchoalveolar lavage fluid;
• clinical assessment to distinguish infection from colonisation;
• identification of additional pathogens; and
• oxygenation-based severity assessment.

The study reinforces the importance of interpreting a positive molecular result in its full clinical context. Detection of P. jirovecii DNA does not always prove that it is causing pneumonia, particularly when another pathogen or diagnosis could explain the illness.

However, the study did not include a comparison group of patients without Pneumocystis pneumonia. It therefore cannot establish the sensitivity or specificity of the proposed workflow or demonstrate that it reliably distinguishes infection from colonisation.

The workflow is best considered a structured clinical proposal requiring prospective validation rather than a new diagnostic standard.

Read the Pneumocystis study in the Journal of Fungi

Overall message

This fortnight’s research shows that progress in fungal diagnostics depends on more than producing new laboratory tests.

The Latin American and Caribbean survey demonstrates an immediate need to make existing tests and antifungal medicines more consistently available. At the same time, carefully developed prediction tools may help clinicians recognise patients who need urgent fungal investigation, especially when characteristic signs are absent.

The externally validated talaromycosis model is the most promising of the new prediction tools, although prospective and international evaluation is still required. The IPA nomogram and Pneumocystis workflow are earlier-stage approaches that should not yet be treated as replacements for established diagnostic methods.

---

This update summarises recently published research for healthcare professionals and others interested in fungal diagnostics. Prediction models support clinical assessment but do not independently confirm or exclude fungal disease. Diagnostic results must be interpreted alongside the patient’s risk factors, symptoms, imaging, microbiology and response to treatment.