Prediction models that reviewers can use.

Risk prediction models developed and validated on your data or on the large databases we work with, from regularised regression to gradient boosting and neural networks, reported to TRIPOD+AI and delivered as a working web calculator.

What it is

A model that predicts, and proves it.

A clinical prediction model estimates an individual's risk of an outcome, death, readmission, complication, response, from things known at the point of decision. Machine learning extends the classical logistic or Cox model with algorithms that can capture non-linear patterns and interactions: penalised (LASSO, ridge, elastic-net) regression, random forests, gradient boosting (XGBoost, LightGBM), and neural networks.

What it is built for

Data, your registry or cohort, or NIS, TriNetX, SEER, UK Biobank, and similar sources.

Methods, regularised regression, random forest, gradient boosting, neural networks; tuning, internal/external validation, calibration, decision curve analysis, SHAP.

Deliverables. TRIPOD / TRIPOD+AI manuscript, reproducible code, and a working web calculator.

How it works

The method, in plain terms.

01

Frame the task

Outcome, horizon, predictors

02

Prepare data

Leakage-safe split

03

Train & tune

Regression to boosting

04

Validate

AUC and calibration

05

Utility

Decision curves, SHAP

06

Report & deploy

TRIPOD+AI, web calculator

How you benefit

Why authors start here.

The project, step by step

How a prediction-model study runs with Cohira.

Start a prediction model →

1

Scope the model

We agree the outcome, the setting, the predictors available at decision time, and whether your data or a large database is the right source.

2

Protocol & sample size

We pre-specify the pipeline and check the data can support the number of predictors without overfitting.

3

Develop

We train and tune candidate algorithms in reproducible Python or R code, with a transparent baseline model for comparison.

4

Validate & explain

Internal and external validation, calibration, decision curve analysis, and SHAP interpretation.

5

Manuscript & calculator

We draft the TRIPOD+AI paper with you and build the web calculator that accompanies it.

6

Submit & revise

We shortlist the right journals, package the submission, and answer reviewers with you to a decision.

Published examples & standards

What prediction-model work looks like in print.

Independent, peer-reviewed examples of what this route can produce.

PLOS ONEComparative model study

Can machine learning improve cardiovascular risk prediction using routine clinical data?

Weng SF, et al. · 2017

Four machine-learning algorithms compared head-to-head with the ACC/AHA risk score in 378,256 patients.

View in journal →
The LancetModel development & validation

An AI-enabled ECG algorithm for identifying patients with atrial fibrillation during sinus rhythm

Attia ZI, et al. · 2019

A neural network trained on 649,931 ECGs and validated on a held-out set, development and validation done properly.

View in journal →
Circulation ResearchIllustrative example

Cardiovascular event prediction by machine learning

Ambale-Venkatesh B, et al. · 2017

Random survival forests on imaging, biomarker, and clinical data from a multi-ethnic cohort outperformed established risk scores.

View in journal →
N Engl J MedReview article

Machine learning in medicine

Rajkomar A, Dean J, Kohane I · 2019

The reference overview of what machine learning can and cannot do in clinical prediction, and the pitfalls reviewers look for.

View in journal →
BMJReporting standard

TRIPOD+AI statement: updated guidance for reporting clinical prediction models

Collins GS, et al. · 2024

The checklist journals now expect for regression and machine-learning prediction models, every Cohira model is written to it.

View in journal →

These are independent publications, shown to illustrate the field and its standards, they are not Cohira's work. We help you develop and publish models of comparable rigor.

Pricing

One fee, from idea to publication.

Original research / prediction model, idea to publication
Protocol, analysis with reproducible code, manuscript, journal shortlist, submission, and reviewer responses.
$1,800per study

Per project, never per author. Journal fees and database access are billed at cost. Your week-by-week timeline is written into the free feasibility note before you pay. Booking more than one study? 10% off the second, 15% off from the third. Acceptance Guarantee & group pricing →

More study types

TriNetXNIS / real-world dataNRD readmissionsGlobal Burden of DiseaseSEER cancer registryUK BiobankMeta-analysisNetwork meta-analysesDiagnostic accuracy meta-analysesNarrative reviewsCross-sectional studiesYour clinical dataSingle stages

Start a conversation

Have an outcome worth predicting?

Tell us the outcome and the data you hold or want to use. You will receive a free feasibility note, design, expected sample, and realistic target journals, before you commit to anything.

Get a free feasibility note See all study types →