Genetic augmentation of a clinical risk model for invasive mould disease

A simulation study of incremental value: data report

Published

November 11, 2026

The question

A validated clinical score (Stanzani et al., J Infect 2019) already predicts 60-day invasive mould disease (IMD) in haematological inpatients with an aROC of 0.85. Candidate host variants in PTX3, CLEC7A (dectin-1), CD209 (DC-SIGN), TLR4 and other genes are repeatedly reported as associated with invasive aspergillosis. This project asks whether adding them to the score would change what a clinician does.

It is framed as a value-of-information question rather than a prediction-improvement exercise:

Under what combinations of effect size, risk-allele frequency and linkage structure would host genetic information change clinical decisions for invasive mould disease in haematological patients, and is the published evidence anywhere near those combinations?

A negative answer, that no realistic genetic architecture moves the decision curve, is a finding and not a failure. The primary endpoint is decision-curve net benefit at the 5% threshold, judged against a minimum clinically important difference fixed before any genetic result exists.

The figures of the draft manuscript and of these pages can be downloaded as 300 dpi PNG files: see the figure folder and its manifest.

Where the project stands

Phase	State
0 Set-up	complete
1 Literature search	complete: 831 unique records screened (PubMed searched 7 July 2026, Embase 8 October 2026)
2 Screening and extraction	round 1 complete, including targeted recovery and supplementary files. Human round-2 checks are reported complete, but the record is not yet in the repository. Direction check of the text-parsed rows closed: 43 rows corrected in all, each with its evidence
3 Parameter derivation	done provisionally for PTX3, CLEC7A (two parameters), TLR4 and CD209; IL10 and IFNG dropped (D14)
4 Baseline	done: refit on the base cohort of 4,078 unique admissions in 1,945 patients (D9, D10), MCID +0.00458 at 5% and +0.00160 at 20%; the published equation, the 4,127 rows and the copula are sensitivity analyses
5 Genetic layer simulation	first results in (addenda 11 to 22 and 24 to 34): exact layer, sampling variability, correlation ensemble, attribution, sensitivity cohorts, donor genotype in the allo-HSCT stratum, Aim 3b, the second CLEC7A parameter, the East Asian check and the tertiary NRI comparison
6 Sensitivity and architecture sweep	Aim 4 sweep done (exact, and with estimated weights) and sensitivity analyses done. Still to do: the pathogen-specificity scenario, and the Aim 3b interaction test on the sensitivity cohorts
7 Write-up	full first draft (Draft manuscript); author list, ethics, data-availability statement and two references still to add

Reading this report

Page	Contents
Draft manuscript	Working draft of the abstract, introduction, methods, results and discussion
Evidence base	What the literature search and extraction produced, and how much of it can be pooled
Genetic panel	The four haplotype terms, linkage structure, allele frequencies, and the haplotype phase check
Baseline model	The fixed clinical equation, the fitted predictor correlation, how it was improved, and how well it is identified
Evidence mapping	How extracted studies become model parameters: the PTX3 h2 definition, the conversion to one scale, and the provisional pooled effect
Decision curves	Exact net benefit for the baseline model across thresholds 1 to 30%, and the re-derived MCID
Pre-specified decisions	The record of design decisions, written before any genetic result
Status log	The full working status document

Source code, data and tests are in the repository.

--- title: "Genetic augmentation of a clinical risk model for invasive mould disease" subtitle: "A simulation study of incremental value: data report" date: today --- ## The question A validated clinical score (Stanzani et al., *J Infect* 2019) already predicts 60-day invasive mould disease (IMD) in haematological inpatients with an aROC of 0.85. Candidate host variants in *PTX3*, *CLEC7A* (dectin-1), *CD209* (DC-SIGN), *TLR4* and other genes are repeatedly reported as associated with invasive aspergillosis. This project asks whether adding them to the score would change what a clinician does. It is framed as a value-of-information question rather than a prediction-improvement exercise: > Under what combinations of effect size, risk-allele frequency and linkage structure would host genetic > information change clinical decisions for invasive mould disease in haematological patients, and is the > published evidence anywhere near those combinations? A negative answer, that no realistic genetic architecture moves the decision curve, is a finding and not a failure. The primary endpoint is decision-curve net benefit at the 5% threshold, judged against a minimum clinically important difference fixed before any genetic result exists. The figures of the draft manuscript and of these pages can be downloaded as 300 dpi PNG files: see the [figure folder](manuscript_figures/README.md) and its [manifest](manuscript_figures/manifest.csv). ## Where the project stands | Phase | State | |---|---| | 0 Set-up | complete | | 1 Literature search | complete: 831 unique records screened (PubMed searched 7 July 2026, Embase 8 October 2026) | | 2 Screening and extraction | round 1 complete, including targeted recovery and supplementary files. Human round-2 checks are reported complete, but the record is not yet in the repository. Direction check of the text-parsed rows closed: 43 rows corrected in all, each with its evidence | | 3 Parameter derivation | done provisionally for PTX3, CLEC7A (two parameters), TLR4 and CD209; IL10 and IFNG dropped (D14) | | 4 Baseline | done: refit on the base cohort of 4,078 unique admissions in 1,945 patients (D9, D10), MCID +0.00458 at 5% and +0.00160 at 20%; the published equation, the 4,127 rows and the copula are sensitivity analyses | | 5 Genetic layer simulation | first results in (addenda 11 to 22 and 24 to 34): exact layer, sampling variability, correlation ensemble, attribution, sensitivity cohorts, donor genotype in the allo-HSCT stratum, Aim 3b, the second CLEC7A parameter, the East Asian check and the tertiary NRI comparison | | 6 Sensitivity and architecture sweep | Aim 4 sweep done (exact, and with estimated weights) and sensitivity analyses done. Still to do: the pathogen-specificity scenario, and the Aim 3b interaction test on the sensitivity cohorts | | 7 Write-up | full first draft ([Draft manuscript](manuscript.qmd)); author list, ethics, data-availability statement and two references still to add | ## Reading this report | Page | Contents | |---|---| | [Draft manuscript](manuscript.qmd) | Working draft of the abstract, introduction, methods, results and discussion | | [Evidence base](evidence.qmd) | What the literature search and extraction produced, and how much of it can be pooled | | [Genetic panel](panel.qmd) | The four haplotype terms, linkage structure, allele frequencies, and the haplotype phase check | | [Baseline model](baseline.qmd) | The fixed clinical equation, the fitted predictor correlation, how it was improved, and how well it is identified | | [Evidence mapping](mapping.qmd) | How extracted studies become model parameters: the PTX3 h2 definition, the conversion to one scale, and the provisional pooled effect | | [Decision curves](decision-curves.qmd) | Exact net benefit for the baseline model across thresholds 1 to 30%, and the re-derived MCID | | [Pre-specified decisions](decisions.qmd) | The record of design decisions, written before any genetic result | | [Status log](status.qmd) | The full working status document | Source code, data and tests are in the [repository](https://github.com/lewis-id-lab/PROSE).