● ResFlow · User guide

ResFlow user guide

From one dataset to manuscript material, this guide lays out the whole workflow and what each step is for. First time using it? Just follow it through once.

1. What it is

ResFlow strings the statistical steps you repeat in clinical research into one line, following the real workflow of "question → analysis → manuscript material". You can use any tool on its own, or — in the Project workbench — paste your data once and run modelling and evaluation in a single pass, with manuscript material assembled automatically. Everything is computed in your own browser; data is never uploaded or stored.

2. Two entry points, pick as needed

🧭 ResFlow home (/en/research)

Use this when you "know you need to analyse but aren't sure which tool". Two ways to find it: the research pipeline map (find by the six pipeline stages) and study-type templates (pick your study type and the tools are listed in order).

🧪 Project workbench (/en/research/project)

Use this when you "have a dataset and want to build and evaluate a prediction model and get manuscript material". Paste data once: modelling → ROC → calibration → DCA → manuscript material, all in one line, no repeated copy-paste. Section 3 below is all about it.

3. Project workbench walkthrough (core)

1

Enter the project dataset

👉 What it is for: Hand your data to the workbench once; every analysis then shares it
Paste your cleaned data into the box: the first row is the column names, each row after that is one record, columns separated by tab or comma (you can copy a header-included range straight from Excel). The outcome column must be coded 0/1 (e.g. 0 = no event, 1 = event). After you click "Save dataset to this project", the workbench auto-detects each column type (numeric / 0-1 / categorical) and stores it in your browser — it does not sync across devices, and nothing is uploaded.
2

One-click baseline table (Table 1)

👉 What it is for: Generate the paper's first table: baseline characteristics by group
Pick a grouping column (e.g. treatment vs control) and click "Generate". The workbench chooses the statistic and test by variable type automatically: mean ± SD for continuous variables, n (%) for categorical variables, with the between-group P value. For a fuller presentation (levels of categorical variables, non-parametric options, etc.), follow the link on the card into the dedicated Table 1 tool.
3

Prediction-model pipeline

👉 What it is for: Build a prediction model and see all three evaluations at once
Choose the outcome column (0/1) and several numeric predictors, then click "Fit model and evaluate". The workbench fits a Logistic regression, reports each factor's OR with 95% CI, and computes a predicted probability for every subject — and that predicted probability is fed automatically into the three evaluations below, with no re-pasting:

• Discrimination · ROC: AUC measures how well the model separates events from non-events; closer to 1 is stronger.

• Calibration: Hosmer-Lemeshow test + calibration slope, measuring whether predicted probabilities are numerically accurate (note: H-L P > 0.05 indicates acceptable calibration).

• Clinical usefulness · DCA: net benefit from the decision curve, measuring whether deciding by this model is actually worthwhile.

Discrimination and calibration are different things: a high AUC does not mean the probabilities are accurate. Look at both, and add DCA for clinical value, to call the evaluation complete.
4

One-click manuscript material

👉 What it is for: Turn the results above into a submittable methods & results draft
This step appears only after a model is fitted (no results, no results text). It writes editable "Statistical methods" and "Results" drafts (numbers taken from your current analysis), pairs them with a TRIPOD reporting checklist (the reporting standard for prediction-model studies, to tick off item by item), and supports copy all or export to Word (.doc) (with baseline table, methods, results, regression table). The draft is a writing aid — before submission, check every number and add your study-design details.

4. Research pipeline map: six stages

On the ResFlow home, the "research pipeline map" groups tools by their order in a study, so you can see what to do at each step:
① Design & planning (choose method, compute sample size) → ② Data & baseline (baseline table, reference interval) → ③ Statistical analysis (group comparison, correlation & regression, survival) → ④ Diagnosis & prediction models (diagnostic performance, ROC, calibration, DCA, nomogram) → ⑤ Reliability & agreement (ICC, Kappa, Cronbach, method comparison) → ⑥ Evidence synthesis (meta-analysis, funnel plot).

5. Study-type templates: match your study

When the workflow is unclear, pick your study type under "study-type templates" and the workbench lays the right tools out in order:
diagnostic accuracy study, prognostic/prediction-model study, cross-sectional study, case-control study, cohort/survival study, scale development & validation, measurement agreement/method study, meta-analysis/systematic review. For example, choose "prognostic/prediction-model study" and it walks you through Design → Baseline → Modelling → Discrimination → Calibration & clinical value → Presentation.

6. FAQ

Q: Will my data be uploaded?
Not by default — all computation runs in your browser and project data is stored locally (localStorage). Only if you sign in to a doctor account and actively click "Save to account / Save as new project" is the project (data and analysis settings) saved to your account, for multi-project management and resuming on another device; otherwise it stays entirely local.

Q: Why don't I see the step-4 manuscript material?
It appears only after a model is successfully fitted in step 3. Confirm the ROC/calibration/DCA results are shown, then scroll down.

Q: Can I use the auto-generated methods/results directly?
As a drafting aid. The numbers come from the modelling data (internal performance, which is optimistic); before submission, check every number, complete the study-design details, and aim to perform external validation.

Q: I need fuller figures and options?
Every result card links to the matching dedicated tool (e.g. full ROC, full calibration curve, full decision curve), where you get richer charts and parameters.

7. Validation & cross-check (external-validation loop)

Every statistical engine has been checked, mostly by the "reduce to a known method" approach — in a special case, the new algorithm should equal a recognised simple one exactly. For example:

· Logistic / Cox regression: coefficients checked against classic datasets (e.g. Cox on the Freireich leukaemia data) and standard implementations;
· Fine-Gray competing risks: reduces to ordinary Cox with no competing events, and to Cox on transformed data with no censoring — consistent from several angles;
· Restricted cubic splines (RCS): basis functions are strictly linear beyond the outer knots (a defining property of splines), with effect = 1 at the reference point;
· Time-dependent AUC: equals the empirical AUC exactly with no censoring;
· VIF: equals 1/(1−R²), infinite under perfect collinearity; PH test: Schoenfeld residuals sum to ≈ 0 at the maximum-likelihood solution (the zero-score property).

Even so, internal performance is still optimistic, so independently cross-check key conclusions. The workbench closes this loop — the three-step R cross-check:

① In step 4, click "Export R code" to copy a script generated for your current model and variables (the RCS tool also has matching rms code);
② Run it in R (the script already includes glm/coxph, pROC, tableone, car::vif, cox.zph, etc.);
③ Use the "R cross-check" checklist below the code to match R's OR/HR, AUC, C-index, PH, VIF, etc. against this tool item by item. Matching direction and magnitude means the implementation agrees.

True external validation should also include re-evaluating the model on an independent population/dataset (not just recomputing on the same data); prediction-model studies should follow the TRIPOD reporting standard.

Get started

With your data ready, run through the Project workbench; to browse by tool first, go to the ResFlow home or the Labs home.