Protocol

Draft and download your review protocol: PRISMA-P for systematic reviews, or a scoping-review protocol. Items prefill from your question and search where available.

What this does

Builds a structured review protocol from the official reporting items, prefilled from your Research Context, and downloads it as an editable Word document you can extend.

When to use

Before screening. A registered protocol (e.g. PROSPERO) reduces bias and is expected by most journals.

How to use

  • Pick systematic review (PRISMA-P) or scoping review
  • Prefill from your project, then complete each item
  • Download the Word file and register it before screening
Moher et al., PRISMA-P 2015; Tricco et al., PRISMA-ScR 2018
Protocol type
Register the finished protocol at PROSPERO (crd.york.ac.uk/prospero) before you start screening.

Full-text review

For studies included at title and abstract: record whether the full text was retrieved and whether it meets eligibility. Feeds the PRISMA flow.

What this does

The second screening stage. It takes the studies you included at title/abstract, tracks full-text retrieval, and records final eligibility with reasons for exclusion.

When to use

After title/abstract screening, before data extraction.

How to use

  • For each study, mark whether the full text was Retrieved or Not retrieved
  • For retrieved reports, mark Eligible or Exclude (with a reason)
  • Sync to PRISMA, then send the eligible studies to extraction
Page et al., PRISMA 2020

PICO Framework

Structure your clinical research question using PICO. Choose PECO for observational/epidemiological studies, or PCC for scoping reviews.

What this does

Structures your research question into Population, Intervention/Exposure, Comparison, and Outcome components.

When to use

Before starting a systematic review, to define a focused and answerable research question.

How to use

  • Select framework variant (PICO, PECO, or PCC)
  • Fill in each component
  • Use the structured question to guide your search strategy and eligibility criteria

Interpreting results

A well-formed PICO question drives every downstream step: search terms, inclusion criteria, data extraction fields, and the structure of your evidence synthesis.

Richardson et al., ACP J Club 1995; Cochrane Handbook Ch. 2
Select Framework
PICO is used for clinical questions in systematic reviews of interventions (RCTs). Best for: "Does treatment X reduce outcome Y in population Z?"
P Population / Problem Who are the participants?
I Intervention What is being studied?
C Comparator / Control What is the alternative?
O Outcome What are the outcomes?
S Study Design What designs are eligible?
PICO Summary
Fill in the fields above to generate your structured research question.

Literature Screening

Upload references from any format, deduplicate, screen titles and abstracts with AI assistance, track PRISMA counts, and send included studies directly to extraction.

What this does

Import references from database exports, remove duplicates, and screen titles/abstracts against your eligibility criteria. AI scoring ranks references by relevance.

When to use

After running your database searches and exporting results as RIS, BibTeX, CSV, or PubMed XML.

How to use

  • Upload reference files (RIS, BibTeX, CSV, XML supported)
  • Set eligibility criteria (PICO, inclusion, exclusion)
  • Use AI scoring to prioritize screening
  • Mark each reference as include, exclude, or unclear
  • Send included references to extraction

Interpreting results

AI scores (0-100) indicate estimated relevance. Always verify AI recommendations manually. PRISMA counts update automatically as you screen.

Key terms

  • RIS, BibTeX, nbib: citation file formats that databases (PubMed, Embase, Scopus) and reference managers (EndNote, Zotero, Mendeley) export. Upload whichever your search produced.
  • Deduplicate: removes the same reference found in more than one database.
  • AI relevance score (0 to 100): an estimate of how well a reference matches your criteria, used to order screening. Always confirm by reading.
Cochrane Handbook Ch. 4; PRISMA 2020 (Page et al., BMJ 2021)
Upload References
Supports RIS, BibTeX (.bib), PubMed XML, CSV, Excel, tab-separated. Duplicates auto-detected.
Upload references into each database below. The count shows how many you have imported from each, and feeds the PRISMA per-database identification counts.
Screening progress
PRISMA counts, live
Records identified 0
Duplicates removed 0
Records screened 0
Excluded 0
Included 0
Eligibility Criteria & AI Settings
Eligibility criteria. These feed the AI screening and are shared with your project.
PICO Framework
Population (P)
Intervention (I)
Comparator (C)
Outcome (O)
Keywords
Bulk: 0 selected 0 references
Upload references to begin screening

Data Extraction

What this does

AI-powered data extraction from published studies.

When to use

After collecting full-text PDFs of included studies.

How to use

  • Pick a review type to load a field preset
  • Add or remove fields so the AI extracts exactly what you need
  • Upload one or more PDFs and extract
  • Review, edit, then export or send to the characteristics table

Interpreting results

Extracted values should always be verified against the source. AI extraction is an aid, not a replacement for manual review.

Cochrane Handbook Ch. 5: Collecting data
1
Review Type
2
Upload PDFs
3
Review & Edit
4
Export
Select review type
Choose the type of systematic review you are conducting.

Scoping Review

Map evidence on a topic, identify gaps, and describe available research.

11 fields

Systematic Review

Critically appraise studies with risk of bias, quality scores, and inclusion criteria.

16 fields

Meta-Analysis (Full)

All fields including effect sizes, CIs, statistical methods, confounders, risk of bias.

24 fields

Quick Meta Extraction

Concise extraction focused on effect sizes, CIs, p-values - ready to paste into the meta-analysis tool.

15 fields
Fields to extract
These are the fields the AI looks for. Start from the preset for your review type, then remove any you do not need or add your own.
Upload study PDFs
Select one or more PDF files. All files are extracted in parallel using AI.

Drop PDF files here

Browse your computer  ·  Multiple files supported

Extracting…
Extracted data
Click any cell to edit. Hover the info icon to view the source quote from the paper.

Meta-analysis

Run pairwise meta-analysis with forest plots, funnel plots, and heterogeneity statistics.

What this does

Inverse-variance meta-analysis with fixed or random effects models.

When to use

After extracting effect sizes (OR, RR, HR, MD, SMD) and confidence intervals from 2+ studies.

How to use

  • Enter study name, effect size, and 95% CI
  • Select effect type and model
  • Run analysis to get pooled estimate, forest plot, funnel plot
  • Check heterogeneity (I2, Q) and publication bias (Egger, Begg)

Interpreting results

I2 < 25% = low heterogeneity, 25-75% = moderate, >75% = high. Egger p < 0.10 suggests publication bias. Check if the pooled CI crosses the null (1 for ratios, 0 for differences).

Key terms

  • Fixed vs random effects: fixed assumes every study estimates one true effect; random allows the true effect to vary between studies. Use random when the studies differ in design or population.
  • Weight: how much each study counts. More precise (usually larger) studies get more weight (inverse-variance weighting).
  • Heterogeneity: how much the results differ beyond chance. Q (Cochran's Q) is the test; I2 is the percent of variation from real differences (under 25 percent low, 25 to 75 moderate, over 75 high); tau2 (and its square root tau) is the estimated between-study variance.
  • DL vs REML: two ways to estimate tau2. DerSimonian-Laird (DL) is the classic method; REML is often preferred for continuous outcomes. They rarely change the conclusion.
  • z statistic: tests whether the pooled effect differs from no effect.
  • Egger and Begg tests: screen for small-study effects and possible publication bias (a p below about 0.10 is a flag, not proof).
  • Subgroup analysis and meta-regression: subgroup analysis pools studies within groups and tests whether the effect differs between them; meta-regression relates the effect to a study-level variable (a moderator, for example mean age).
DerSimonian & Laird 1986; Higgins et al., Cochrane Handbook Ch. 10
Analysis configuration
Select your framework, study design, and data type before entering study data.

1. Framework

2. Study design

3. Available data

Binary and Continuous take raw per-group data (events/total, or mean/SD/n) and compute the effect for you. Pre-computed takes an already-calculated HR, OR or RR with its 95% CI: use it for time-to-event (HR), or when a study only reports the summary estimate. OR and RR appear in both because you can either enter raw 2x2 counts (Binary) or a pre-calculated ratio (Pre-computed).

4. Select outputs

Study data
Enter effect sizes and 95% confidence intervals for each study. Values must be on the original scale (not log-transformed).
Author(s) Year Effect Size 95% CI Lower 95% CI Upper Subgroup
Effect measure
Model
Heterogeneity estimator
 
Minimum 2 studies required

PRISMA 2020 Flow Diagram

Generate a publication-ready PRISMA 2020 flow diagram for your systematic review.

What this does

PRISMA 2020 flow diagram generator.

When to use

To report the study selection process in a systematic review.

How to use

  • Enter counts for each stage
  • Provide exclusion reasons
  • Select PRISMA variant (2009 or 2020)
  • Export as SVG or PNG

Interpreting results

The diagram shows how many records were found, screened, excluded, and included. Transparent reporting of the selection process.

Page et al., BMJ 2021 (PRISMA 2020)
PRISMA 2020 Flow Diagram Generator
Choose the diagram type that matches your review, fill in the numbers, and generate a publication-ready flow chart.
NEW SR
Standard
Databases & registers only. Single identification stream.
NEW SR
+ Other Methods
Databases/registers + grey literature, websites, citation searching.
UPDATED SR
Standard Update
Previous version studies + new database search results combined.
Prev
UPDATED SR
Full Update
Previous studies + databases + other methods. Most comprehensive.
Identification - Databases & Registers
Tick each database searched and enter record count:
Screening
Eligibility
Included
Chart Style & Colours
Presets
#ffffff
#1e40af
#fff5f5
#dc2626
#eff6ff
#1e40af
#1e40af
#f8fafc
#0f172a
#1e40af
Override box labels (optional)
2px
2px
13px
Chart style: Publication SVG renders a print-ready vector diagram

Risk of Bias Assessment

Assess study quality using validated tools: RoB 2, NOS, and AXIS for multiple study designs.

What this does

Risk of bias assessment using standard tools (RoB 2, ROBINS-I, NOS, AXIS, JBI).

When to use

During quality assessment of included studies.

How to use

  • Select the appropriate tool for your study design
  • Add studies and rate each domain
  • View the traffic light summary
  • Export as CSV, SVG, or PNG

Interpreting results

Low = low risk. Some concerns = potential issues. High = likely bias affecting results. Overall judgement follows the worst domain rating.

Key terms

  • Domain: a specific source of bias you judge (for example how participants were randomised, or how the outcome was measured).
  • The tools: RoB 2 for randomised trials (5 domains); ROBINS-I for non-randomised studies of interventions (7 domains); NOS (Newcastle-Ottawa Scale) for cohort and case-control studies; AXIS for cross-sectional studies; JBI (Joanna Briggs Institute) checklists for several designs.
  • Judgements: RoB 2 uses Low, Some concerns, High. ROBINS-I uses Low, Moderate, Serious, Critical. The overall rating follows the worst domain.
  • Traffic light summary: the coloured grid of each study's rating per domain (green low, yellow some concerns, red high).
Sterne et al., BMJ 2019 (RoB 2); Sterne et al., BMJ 2016 (ROBINS-I)
Risk of Bias Assessment
Choose the appraisal tool that matches your study design, then add studies manually, paste from a spreadsheet, or upload a file.
Spreadsheet format
Citation for this tool

Leave-One-Out Sensitivity Analysis

Assess robustness by systematically removing one study at a time. Identifies influential studies.

What this does

Re-runs the meta-analysis k times, each time removing one study. Shows how each study influences the pooled estimate.

When to use

After running a meta-analysis, to check if the result depends on any single study.

How to use

  • Run a meta-analysis first
  • Open Leave-One-Out and click Run
  • Review the table and forest plot showing pooled estimates with each study removed

Interpreting results

If removing one study changes the pooled estimate direction or significance, that study is influential. Consider subgroup analysis or investigate why it differs.

Key terms

  • Leave-one-out: re-runs the pooled analysis with each study removed in turn, to see how much any single study drives the result.
  • Influential: a study is flagged when removing it shifts the pooled estimate enough to matter (changing its direction or whether it stays statistically significant).
Viechtbauer & Cheung, Res Synth Methods 2010
Study data
Import from Meta-analysis or enter data manually. Effect measure and model settings are shared with the Analysis page.
Author(s)YearEffect Size95% CI Lower95% CI Upper
Effect measure
Model
Minimum 3 studies required

Network Meta-Analysis

Visualise your treatment network, compute indirect comparisons, rank treatments by P-score, and generate a league table. Based on frequentist NMA methodology.

What this does

Network meta-analysis comparing multiple treatments simultaneously.

When to use

When 3+ treatments have been compared across studies, with both direct and indirect evidence.

How to use

  • Add treatments
  • Enter pairwise comparisons (studies, effect, SE)
  • Generate network diagram
  • Review league table, rankings, and P-scores

Interpreting results

P-score near 1 = likely best treatment. Check network connectivity and consistency. Inconsistency suggests the indirect evidence disagrees with direct evidence.

Key terms

  • Direct vs indirect evidence: direct compares two treatments tested head to head; indirect compares them through a shared comparator (A vs B inferred from A vs C and B vs C).
  • Bucher method: the calculation that combines direct comparisons into an indirect one.
  • League table: a grid of every treatment against every other, read as row versus column.
  • P-score: a 0 to 1 score for how likely each treatment is to be among the best (near 1 is best). It is the frequentist version of SUCRA.
  • Consistency: agreement between the direct and indirect evidence in a closed loop of the network. Disagreement is called inconsistency.
  • SE (standard error): the precision of a comparison's effect estimate, smaller is more precise.
Rucker & Schwarzer, Stat Med 2015; Salanti, Ann Intern Med 2011
Treatments
Direct Comparisons
Studies and effect estimates for each pair:
Network Diagram

Add treatments and click Generate

League Table of All Pairwise Comparisons
Read: row treatment vs column treatment. Values show effect estimate (95% CI). Diagonal = reference. Green = favours row; red = favours column. Generate the network diagram first.
Generate the network diagram first, then the league table will appear here.
Consistency Assessment
Checks whether direct and indirect evidence agree for each closed loop in the network.
Generate the network first.
Treatment Rankings - P-score
P-score (frequentist analogue of SUCRA): probability that a treatment is better than any randomly chosen treatment. Choose which direction of the effect counts as the better outcome; the data you enter do not change, only the ranking direction.
Better outcome
Ranking Table
Generate the network first.
NMA Methodology
Frequentist NMA uses a graph-theoretic approach (Rücker 2012) implemented via the hat matrix of the network. The treatment effects are estimated by solving the system of equations that relate direct and indirect evidence.
Network geometry
Treatments (nodes): k
Direct comparisons (edges): m
Degrees of freedom for inconsistency: m − k + 1
Total studies: n
P-score (Rücker & Schwarzer 2015)
P-score(A) = mean over all B≠A of P(μ_A > μ_B)
≈ Φ( μˆ_AB / SE(μˆ_AB) )
where μˆ_AB is the NMA estimate for A vs B
Indirect estimate (Bucher method for single loop)
θ_AC_indirect = θ_AB_direct − θ_CB_direct
SE² = SE²_AB + SE²_CB
Inconsistency factor IF = θ_direct − θ_indirect
Bayesian NMA
Bayesian NMA (WinBUGS / R: gemtc, BUGSnet) provides posterior distributions for treatment effects under a random-effects model with informative or non-informative priors. This cannot be run in-browser. Use the R code template below and run in R with the gemtc package:
library(gemtc)
network <- mtc.network(data.re = your_data)
model <- mtc.model(network, type="consistency", linearModel="random")
results <- mtc.run(model, n.adapt=5000, n.iter=20000, thin=10)
summary(results)
rank.probability(results)

References: Rücker G. Stat Med. 2012;31:1236–1251. | Rücker G, Schwarzer G. BMC Med Res Methodol. 2015;15:58. | Salanti G et al. Ann Intern Med. 2011;154:544–553. | Dias S et al. Stat Med. 2013;32:739–776.

Forest Plot Builder

Build a publication-ready forest plot from any effect estimates - no raw data needed. Enter your studies manually or paste from a spreadsheet.

What this does

Standalone forest and funnel plot builder with full style control.

When to use

When you need publication-ready plots from pre-computed effect sizes.

How to use

  • Enter study data (name, effect, CI, weight)
  • Select effect type and adjust styles
  • Download as SVG or PNG

Interpreting results

Forest plot: each line = one study CI. Diamond = pooled estimate. Funnel plot: asymmetry suggests publication bias.

Lewis & Clarke, BMJ 2001
Plot Settings
Plot Title
Effect Measure Label
Null line at
Log scale
Point colour
Diamond colour
Font size
Style
X-axis min
X-axis max
CI line width
Pooled label
Background
Studies
Paste directly from Excel/Sheets (columns: Study | Effect | CI Lower | CI Upper | Weight%)
Study / Label Effect CI Lower CI Upper Weight % Group
Forest Plot

Effect Size Calculator

Convert between effect measures, calculate NNT/NNH, and transform effect sizes for meta-analysis.

What this does

Three calculators: (1) NNT and absolute risk from an effect estimate plus a baseline risk; (2) convert between effect measures (OR, RR, RD, SMD); (3) recover the standard error and variance from a 95% confidence interval.

When to use

To translate an effect into plain numbers (how many people benefit), to convert a measure into the one your analysis needs, or to get the SE a meta-analysis requires from a published CI.

How to use

  • Pick a tab and enter the values; results update as you type.
  • For NNT, enter the effect and the baseline (control) risk.
  • Use Sample on any tab for a worked example.

Key terms

  • CER (control event rate): the outcome rate in the control or untreated group, i.e. the baseline risk.
  • EER (experimental event rate): the outcome rate in the treated or exposed group.
  • RR (risk ratio): EER divided by CER. Below 1 treatment lowers risk; above 1 it raises it.
  • OR (odds ratio): the odds of the outcome in the treated group divided by the odds in the control group.
  • HR (hazard ratio): the ratio of event rates over time, used in survival analysis.
  • RD / ARR (risk difference, absolute risk reduction): CER minus EER, the absolute change in risk. A negative value is an absolute risk increase (ARI).
  • RRR (relative risk reduction): the share of the baseline risk removed by treatment, equal to 1 minus RR.
  • NNT (number needed to treat): how many people must be treated for one extra good outcome, equal to 1 divided by ARR. NNH (number needed to harm) is the same idea for one extra bad outcome.
  • SMD (standardised mean difference) and Cohen's d: a mean difference in standard-deviation units (about 0.2 small, 0.5 medium, 0.8 large).
  • SE (standard error) and 95% CI (confidence interval): how precise an estimate is.

Interpreting results

For ratios (OR, RR, HR) a value of 1 means no difference between groups. A smaller NNT means a larger benefit per person treated.

Borenstein et al., Introduction to Meta-Analysis 2009
NNT / Absolute Risk Calculator
Effect measure
Effect value
Baseline risk (CER) %
Effect Measure Converter
Input measure
Input value
Baseline risk % (for OR→RR)
SE / Variance from 95% CI
Effect size
95% CI Lower
95% CI Upper
Log scale?

CONSORT Flow Diagram

Generate a CONSORT 2010 flow diagram for your randomised controlled trial.

What this does

Generates a CONSORT 2010-compliant flow diagram showing participant flow through enrollment, allocation, follow-up, and analysis.

When to use

When reporting results of a randomised controlled trial.

How to use

  • Enter participant counts at each stage
  • Specify exclusion and loss-to-follow-up reasons
  • Export as SVG or PNG

Interpreting results

The diagram must account for every participant from enrollment to analysis. Discrepancies between allocated and analysed numbers should be explained.

Schulz et al., Ann Intern Med 2010 (CONSORT 2010)
Enrollment
Trial arms
Chart Style & Colours
Presets
#ffffff
#1e40af
#fff5f5
#dc2626
#1e40af
#f8fafc
#0f172a

GRADE Evidence Table

Rate the certainty of evidence for each outcome using the GRADE approach.

What this does

GRADE certainty of evidence assessment for each outcome.

When to use

After completing meta-analysis and risk of bias assessment.

How to use

  • Add an outcome
  • Set study design (RCT starts at High, Observational at Low)
  • Rate 5 downgrade and 3 upgrade criteria
  • Review the final certainty rating

Interpreting results

High = very confident. Moderate = moderately confident. Low = limited confidence. Very Low = very little confidence. Each serious downgrade drops one level.

Key terms

  • Certainty of evidence: how much the estimate for an outcome can be trusted, rated High, Moderate, Low, or Very Low.
  • Five reasons to downgrade: risk of bias (flaws in how studies were run), inconsistency (results differ across studies), indirectness (the evidence does not match your exact question, population, or outcome), imprecision (wide confidence intervals or few events), and publication bias (missing studies).
  • Three reasons to upgrade (for observational evidence): a large effect, a dose-response gradient, and plausible confounding that would only work against the observed effect.
  • RCTs start at High and observational studies at Low; each serious concern moves the rating one level down (or up).
Guyatt et al., J Clin Epidemiol 2011; GRADE Working Group

Characteristics of Included Studies

Build the "Table 1" summary of included studies for your systematic or scoping review. Add studies as rows and define columns based on your review type.

What this does

Builds the characteristics table summarizing key attributes of each included study. Column presets match common review types.

When to use

After data extraction, to create the summary table required in every systematic review.

How to use

  • Select a column preset (scoping, systematic, meta-analysis, RCT) or customize columns
  • Add studies as rows
  • Fill in each cell
  • Export as CSV or copy to your manuscript

Interpreting results

This table provides readers with an overview of included studies. Ensure consistent reporting across all rows. Missing data should be marked explicitly.

Cochrane Handbook Ch. 5; PRISMA 2020 Checklist Item 18
Column preset
Columns (drag to reorder, click × to remove)
Studies

Randomization Tool

Generate allocation sequences for clinical trials. Supports simple, block, and stratified randomization.

What this does

Generates randomization sequences for clinical trials using simple, block, or stratified methods with configurable allocation ratios.

When to use

Before starting participant enrollment in a clinical trial.

How to use

  • Set the number of participants and groups
  • Choose randomization method (simple, block, stratified)
  • Set block size if using block randomization
  • Generate and export the allocation sequence

Interpreting results

Block randomization ensures balanced group sizes at regular intervals. Stratified randomization ensures balance across important prognostic factors. Keep the sequence concealed from recruiters.

Schulz & Grimes, Lancet 2002; CONSORT 2010
Setup
Number of arms (2–4)
Allocation ratio
Total participants
Randomization type
Seed (for reproducibility)

Trend Analysis & Segmented Regression

Joinpoint regression with automatic BIC-based model selection. Calculate APC (Annual Percent Change) and AAPC (Average Annual Percent Change) with Delta Method confidence intervals.

What this does

Joinpoint regression with BIC/AIC model selection. Calculates APC and AAPC.

When to use

For time-series data (e.g. annual incidence rates, mortality trends).

How to use

  • Paste Year and Value columns (supports multiple outcome columns)
  • Select scale (log for rates, linear for counts)
  • Choose auto or manual joinpoint selection
  • Review APC per segment and overall AAPC

Interpreting results

APC = Annual Percent Change within a segment. AAPC = weighted average across all segments. Negative APC = decreasing trend. Joinpoints mark where the trend direction changes.

Key terms

  • Joinpoint: a year where the trend changes slope. The tool fits straight segments joined at these points.
  • APC (annual percent change): the yearly percent change within one segment. AAPC (average APC): a weighted average across all segments.
  • Log-linear vs linear scale: pick log-linear for rates, proportions, or percentages (a constant percent change), and linear for counts or values that change by a fixed amount.
  • BIC / AIC / AICc: scores that choose how many joinpoints to keep, balancing fit against simplicity. Leave on the default unless you want to set the number yourself.
  • Weighted least squares: the fitting method, giving more weight to more precise points; the AAPC confidence interval uses the delta method.
Kim et al., Stat Med 2000; NCI Joinpoint Regression Program
How to format your data
Three ways to use this tool. Click "Load example" on any one to see it run.
1. One outcome (Year + value)
Paste two columns into the Data Input box below, then click Run Analysis.
Year  Rate
2005  8.2
2006  9.1
2007  10.5
 ...
2. Several outcomes (wide)
Year in the first column, each outcome in its own column. Paste below; a column picker appears so you choose which outcomes to plot.
Year  Incidence  Mortality
2005  8.2  3.1
2006  9.1  3.4
 ...
3. Factor variables (long format)
One row per year per group. Use "Upload data with factor variables" below: pick the Year column, the outcome (or count ÷ population), and the factor columns to split by.
year,sex,onset,count,population
1990,Female,Early,241334,1081823027
1990,Male,Early,264921,1109228616
 ...
Data Input
Paste from Excel: first column = Year, remaining columns = outcome values. Header row is auto-detected. For data with factor variables (sex, age group, type), use the upload box below instead.
Upload data with factor variables
+
Model Selection
Joinpoint selection
Maximum joinpoints to test
Information criterion
Scale
Min years between joinpoints
Paste data and click Run Analysis
Chart Settings
Chart Title
Y-axis Label
X-axis (blank = auto)
Y-axis (blank = auto)
Data colour
Segment colour
Width (px)
Height (px)
Common R sizes at 96 dpi — 7×5 in: 672×480  ·  8×6 in: 768×576  ·  10×7 in: 960×672

Statistical Analysis

Comprehensive statistics center - descriptive, comparison, correlation, regression, survival, diagnostic accuracy, prediction validation, and causal methods. Paste data from Excel, upload CSV, or enter manually. All in-browser calculations use transparent, established formulas. Code templates available for R and Python.

What this does

Statistical tests: t-test, chi-square, ANOVA, correlation, regression, Kaplan-Meier, ROC.

When to use

For individual study analysis or exploratory data analysis.

How to use

  • Select the test type
  • Paste or enter your data
  • Run the analysis
  • Review test statistics and p-values

Interpreting results

p < 0.05 is the conventional threshold for statistical significance. Always consider effect size alongside p-values. Check assumptions (normality, independence) before interpreting.

Key terms by tab

Comparison: a t-test compares two means (Welch does not assume equal variances); ANOVA compares three or more means; Mann-Whitney, Wilcoxon signed-rank, and Kruskal-Wallis are rank-based versions for non-normal data; chi-square, Fisher's exact, and McNemar's test associations in count tables; p-value corrections (Bonferroni, Holm, Benjamini-Hochberg) adjust for testing many hypotheses.

Correlation: Pearson (linear) and Spearman or Kendall's tau (rank-based) measure association from -1 to 1. ICC (intraclass correlation) measures agreement between raters.

Regression: OLS for continuous outcomes; logistic gives an OR (odds ratio); log-binomial and modified Poisson give an RR (risk ratio); Poisson and negative binomial give an IRR (incidence rate ratio, for counts; negative binomial adds theta for overdispersion); Cox gives an HR (hazard ratio); ordinal and multinomial handle ordered or unordered categories. beta is the coefficient, SE its standard error, R2 the variance explained.

Survival: Kaplan-Meier estimates survival over time; censored means the event had not happened by last follow-up; the log-rank test compares groups.

Diagnostic and ROC: sensitivity (true positives found), specificity (true negatives found), PPV/NPV (predictive values), DOR (diagnostic odds ratio), AUC (area under the ROC curve, 0.5 is chance and 1.0 is perfect), and the Youden index (best cut-off).

Prediction: the confusion matrix and accuracy, precision, recall, F1 summarise classification; the Brier score measures probability accuracy (lower is better); calibration checks predicted against observed risk; k-fold cross-validation tests on held-out data.

Causal and clustering: the Mann-Kendall test detects a monotonic trend; k-means groups observations into clusters (WCSS is within-cluster spread, used by the elbow plot).

Various: Welch 1947, Pearson 1900, Kaplan & Meier 1958
Data Manager
Upload or paste your dataset to enable Fill from data on all analyses below. Drag & drop a file here too.
Continuous Data - Summary Statistics
Paste one column of numeric values (one per row). For group comparison, use two tab-separated columns.
Results
Enter data and click Calculate
Frequency Table
Paste a column of categorical values (one per row, e.g. Male/Female/Other)
Frequency Results
Enter categories and click Build Table
Cross-Tabulation
Paste two tab-separated columns (var1[tab]var2 per row). First row can be a header.
Cross-Tab Results
Enter data and click Build Cross-Tab
Missing Data Summary
Paste data with headers (tab-separated columns). Blanks, "NA", "N/A", "." treated as missing.
Enter data and click Analyze
Prevalence / Rate Calculator
Cases
Population
Rate per
Test type
Test Options
Test type
Hypothesis
α level
Group 1
Group 2
Results
Enter data and click Run T-test
Method
Variable X
Variable Y
Results
Enter data and click Calculate
Scatter Plot
Intraclass Correlation Coefficient (ICC)
Paste a matrix: rows = subjects, columns = raters (tab-separated). First row optional header.
ICC Results
Enter rater matrix and click Calculate ICC
Correlation Matrix
Paste multiple columns (tab-separated, first row = variable names). Computes all pairwise Pearson correlations.
Enter data and click Build Matrix
Regression type
Linear Regression (OLS)
Outcome (Y) - one column
Predictor(s) (X) - tab-separated columns, one row per observation
Stratify by (column header in X)
Results
Enter data and click Run Regression
Regression Plot
Survival / Time-to-Event Data
Paste columns: time (numeric), event (0/1, 1=event), optional group. Use the group column to compare 1 or 2 strata. First row = headers.
Summary
Enter data and click Run analysis
Diagnostic Accuracy - 2×2 Table
Enter TP, FP, FN, TN counts
Disease + Disease −
Test +
Test −
Diagnostic Results
Enter counts and click Calculate
ROC Curve Builder
Paste two columns: predicted probability (0–1) and true label (0/1). First row = headers.
ROC Summary
Enter data and click Plot ROC
ROC Curve
Confusion Matrix from Predictions
Paste two columns: predicted (0/1 or class labels) and actual (0/1 or class labels). First row = headers.
Classification Metrics
Enter data and click Compute Metrics
Brier Score & Calibration
Paste two columns: predicted probability (0–1) and actual outcome (0/1). First row = headers.
Brier Score Results
Enter data and click Compute
Calibration Plot
K-Fold Cross-Validation Splitter
Total observations (N)
Number of folds (k)
Cross-Validation Code Templates
Complete pipelines for k-fold CV in R and Python (sklearn + caret)
Mann-Kendall trend test and K-means cluster analysis, computed in-browser.
Mann-Kendall Trend Test (In-Browser)
Paste a time-series (one value per row, in chronological order)
Mann-Kendall Results
Enter time series and click Run Mann-Kendall
K-Means Cluster Analysis (In-Browser)
Paste numeric data (tab-separated columns, one observation per row). First row optional headers.
K (clusters)
K-Means Results
Enter data and click Run K-Means
Cluster / Elbow Plot

AI Data Advisor

Upload your dataset and describe your research objectives. Our AI will analyse your data structure, suggest appropriate statistical analyses and figures, and generate a ready-to-use prompt you can take to any AI assistant (Claude, ChatGPT, etc.) to instantly get complete R and Python code - without having to explain your data yourself.

What this does

Analyses your dataset structure and research objectives. Suggests statistical tests, figures, and generates a complete prompt for R/Python code generation.

When to use

When you have a dataset and need guidance on which analyses to run, or want ready-made code for R or Python.

How to use

  • Upload a CSV or Excel file
  • Describe your research objectives
  • Review suggested analyses and figures
  • Copy the generated prompt into any AI tool for complete code

Interpreting results

Suggestions are based on variable types and research context. Always verify that suggested tests match your study design assumptions. The generated prompt includes your actual column names and data types.

Kovex AI Advisor uses Gemini for data structure analysis
Step 1 - Upload or Paste Your Data
Upload a CSV file or paste your data directly. Headers are required. The AI will read your variable names and first rows to understand your dataset.
Click to upload CSV
or drag and drop · CSV files only
Recommended for best AI results: organise your data in this template, then upload it. Not required - any tidy CSV works.
- or paste data directly -
Step 2 - Describe Your Research Objectives (optional but recommended)
Upload data to get started
AI analysis results will appear here

Methods, Transparency & Validation

Complete documentation of every statistical formula, algorithm, and decision rule used by Kovex. Each section includes the mathematical specification, academic references, and downloadable R replication code so you can independently verify our results.

What this does

Documents every formula, algorithm, and decision rule used across all Kovex tools. Includes R replication code for independent verification.

When to use

When writing your methods section, responding to peer review, or verifying that Kovex calculations match your expectations.

How to use

  • Browse by tool or statistical method
  • Review the mathematical specification
  • Download R code to replicate any calculation independently
  • Cite the listed references in your manuscript

Interpreting results

This page is reference documentation. Use it to confirm which formulas Kovex applies and to write accurate methods descriptions for publication.

All references are listed per method within each section
Open-source transparency
Every computation in Kovex is documented below with the exact formula, the academic source, and a downloadable R script you can run locally to reproduce and verify our results. We believe research tools should be auditable.
Kovex v1.0 - Last methodology update: May 2026

Meta-analysis - Pooled Effect & Heterogeneity

Kovex implements both fixed-effect (Inverse-Variance) and random-effects (DerSimonian-Laird) models for pooling study-level effect estimates.

Study Weights - Fixed Effect (Inverse-Variance)
w_i = 1 / v_i
where v_i = variance of study i effect estimate
Between-Study Variance - DerSimonian-Laird tau-squared
Q = Sum[ w_i * (theta_i - theta_bar)^2 ] (Cochran's Q statistic)
C = Sum(w_i) - Sum(w_i^2) / Sum(w_i)
tau^2 = max( 0, (Q - (k-1)) / C )
Study Weights - Random Effects
w_i* = 1 / (v_i + tau^2)
Pooled Effect Estimate
theta_hat = Sum(w_i* * theta_i) / Sum(w_i*)
SE(theta_hat) = sqrt(1 / Sum(w_i*))
95% CI: theta_hat +/- 1.96 * SE(theta_hat)
Ratio Measures (OR, RR, HR)
Analysis on log scale: theta_i = ln(effect)
Back-transformed: exp(theta_hat), exp(CI bounds)
Heterogeneity Statistics
I^2 = max(0, 100% * (Q - (k-1)) / Q)
tau = sqrt(tau^2)
H^2 = Q / (k-1)
p-value: chi-squared distribution with (k-1) df
Prediction Interval (95%)
PI = theta_hat +/- t(0.975, k-2) * sqrt(tau^2 + SE(theta_hat)^2)

References: DerSimonian R, Laird N. Control Clin Trials. 1986;7:177-188. | Higgins JPT, Thompson SG. Stat Med. 2002;21:1539-1558. | Borenstein M et al. Introduction to Meta-Analysis. Wiley; 2009.

R Replication Code
# Replicate Kovex meta-analysis in R
library(meta)

# Example: generic inverse-variance input
study   <- c("Study A","Study B","Study C","Study D","Study E")
effect  <- c(0.42, 0.58, 0.31, 0.67, 0.45)   # log(OR) or SMD
se      <- c(0.18, 0.22, 0.15, 0.25, 0.20)

m <- metagen(TE = effect, seTE = se, studlab = study,
             sm = "OR", method.tau = "DL",
             fixed = TRUE, random = TRUE)
summary(m)
forest(m)
funnel(m)

Leave-one-out Sensitivity Analysis

Each study is systematically removed and the meta-analysis re-run on the remaining k-1 studies to identify influential studies.

For each study i (i = 1 ... k):
theta_hat(-i) = pooled estimate omitting study i
tau^2(-i) = between-study variance omitting study i
I^2(-i) = heterogeneity omitting study i

Reference: Viechtbauer W, Cheung MWL. Res Synth Methods. 2010;1:112-125.

R Replication Code
# Leave-one-out in R
library(metafor)

yi <- c(0.42, 0.58, 0.31, 0.67, 0.45)
sei <- c(0.18, 0.22, 0.15, 0.25, 0.20)

res <- rma(yi = yi, sei = sei, method = "DL")
loo <- leave1out(res)
print(loo)
forest(loo)

Effect Size Calculator - NNT, ARR, RR, OR, SMD

From 2x2 Table (a, b, c, d)
EER = a / (a+b) CER = c / (c+d)
RR = EER / CER
OR = (a*d) / (b*c)
RD = EER - CER
ARR = CER - EER
NNT = 1 / |ARR|
Confidence Intervals
SE(ln OR) = sqrt(1/a + 1/b + 1/c + 1/d)
SE(ln RR) = sqrt(1/a - 1/(a+b) + 1/c - 1/(c+d))
SE(RD) = sqrt(EER*(1-EER)/(a+b) + CER*(1-CER)/(c+d))
SMD (Standardised Mean Difference)
Cohen's d = (M1 - M2) / SD_pooled
SD_pooled = sqrt(((n1-1)*SD1^2 + (n2-1)*SD2^2) / (n1+n2-2))
Hedges' g = d * (1 - 3/(4*(n1+n2)-9))

References: Borenstein M et al. Wiley; 2009. | Zhang J, Yu KF. JAMA. 1998;280:1690-1691.

R Replication Code
# Effect size calculations in R
library(epitools)
library(meta)

# 2x2 table: a=15, b=85, c=30, d=70
tab <- matrix(c(15, 85, 30, 70), nrow=2, byrow=TRUE)
oddsratio.wald(tab)
riskratio.wald(tab)

# SMD (Hedges' g)
metacont(n.e=50, mean.e=12.3, sd.e=3.1,
         n.c=50, mean.c=10.1, sd.c=2.9, sm="SMD")

Funnel Plot & Publication Bias

x-axis: theta_i (study effect)
y-axis: SE(theta_i) inverted
Pseudo-95% CI: theta_hat +/- 1.96 * SE

Asymmetry suggests publication bias but may also reflect heterogeneity or small-study effects. Formal tests require 10+ studies.

References: Sterne JAC et al. BMJ. 2011;343:d4002. | Egger M et al. BMJ. 1997;315:629-634.

Joinpoint Trend Analysis - APC & AAPC

Kovex fits piecewise log-linear regression to time-series rate data, identifying points where the trend changes direction (joinpoints).

Model
ln(rate) = beta_0 + beta_1*year + delta_1*(year - tau_1)+ + delta_2*(year - tau_2)+ + ...
where (x)+ = max(0, x) and tau_j are joinpoints
Annual Percent Change (APC)
APC = 100 * (exp(slope) - 1)
95% CI: 100 * (exp(slope +/- 1.96*SE) - 1)
Average APC (AAPC)
AAPC = 100 * (exp(weighted avg of segment slopes) - 1)
Weights = segment length / total period

References: Kim HJ et al. Stat Med. 2000;19:335-351. | Muggeo VMR. Stat Med. 2003;22:3055-3071. | Clegg LX et al. Stat Med. 2009;28:3670-3682.

R Replication Code
# Joinpoint trend analysis in R
library(segmented)

year <- 2000:2020
rate <- c(12.1,12.4,12.8,13.5,14.2,14.8,15.1,15.0,14.6,14.2,
          13.8,13.5,13.9,14.5,15.2,16.1,17.0,17.8,18.2,18.5,18.7)

fit <- lm(log(rate) ~ year)
seg <- segmented(fit, seg.Z = ~year, npsi = 2)
summary(seg)
slope(seg)  # APC per segment
plot(seg)

Network Meta-Analysis - P-score & League Table

Kovex implements frequentist NMA with Bucher indirect comparisons, league table generation, node-splitting consistency assessment, and P-score ranking.

Indirect Comparison (Bucher Method)
theta_AB(indirect) = theta_AC - theta_BC
var(theta_AB) = var(theta_AC) + var(theta_BC)
P-score (Frequentist SUCRA Analogue)
P-score_i = (1/(k-1)) * Sum[ Phi(theta_ij / sqrt(var_ij)) ] for all j != i
where Phi() = standard normal CDF

References: Salanti G. Res Synth Methods. 2012;3:80-97. | Rucker G, Schwarzer G. Res Synth Methods. 2015;6:260-267. | Bucher HC et al. J Clin Epidemiol. 1997;50:683-691.

R Replication Code
# NMA in R using netmeta
library(netmeta)

treat1 <- c("A","A","B")
treat2 <- c("B","C","C")
TE     <- c(0.5, 0.8, 0.3)
seTE   <- c(0.2, 0.25, 0.22)

net <- netmeta(TE, seTE, treat1, treat2, sm="OR")
summary(net)
netrank(net)        # P-scores
netheat(net)        # Consistency
netleague(net)      # League table

Statistics Suite - Tests, Correlation, Survival, ROC

Two-sample t-test
t = (M1 - M2) / sqrt(s1^2/n1 + s2^2/n2)
df (Welch) = (s1^2/n1 + s2^2/n2)^2 / ((s1^2/n1)^2/(n1-1) + (s2^2/n2)^2/(n2-1))
Chi-squared Test
chi^2 = Sum[ (O_ij - E_ij)^2 / E_ij ]
E_ij = (row_total_i * col_total_j) / grand_total
df = (rows-1) * (cols-1)
Pearson Correlation
r = Sum[(x_i - x_bar)(y_i - y_bar)] / sqrt(Sum[(x_i - x_bar)^2] * Sum[(y_i - y_bar)^2])
t = r * sqrt((n-2)/(1-r^2)), df = n-2
Kaplan-Meier Survival
S(t) = Product[ (1 - d_j / n_j) ] for all t_j <= t
SE(S) = S(t) * sqrt(Sum[ d_j / (n_j*(n_j - d_j)) ]) (Greenwood)
ROC / AUC (Trapezoidal)
AUC = Sum[ (FPR_{i+1} - FPR_i) * (TPR_{i+1} + TPR_i) / 2 ]
SE(AUC) via DeLong method
ANOVA (One-way)
F = MS_between / MS_within
MS_between = SS_between / (k-1)
MS_within = SS_within / (N-k)

References: Kaplan EL, Meier P. J Am Stat Assoc. 1958;53:457-481. | DeLong ER et al. Biometrics. 1988;44:837-845. | Welch BL. Biometrika. 1947;34:28-35.

R Replication Code
# Statistics suite replication
# t-test
t.test(x, y, var.equal = FALSE)

# Chi-squared
chisq.test(matrix(c(a,b,c,d), nrow=2))

# Correlation
cor.test(x, y, method = "pearson")

# Kaplan-Meier
library(survival)
fit <- survfit(Surv(time, event) ~ group, data = df)
summary(fit)

# ROC / AUC
library(pROC)
roc_obj <- roc(label, prob)
auc(roc_obj)
ci.auc(roc_obj)

Risk of Bias Assessment Tools

RoB 2 (Randomised Trials)

Five domains: randomization process, deviations from interventions, missing outcome data, outcome measurement, selection of reported results. Each: Low / Some Concerns / High. Overall = most severe domain.

ROBINS-I (Non-randomised Studies of Interventions)

Seven domains: confounding, selection, classification of interventions, deviations, missing data, measurement, reported results. Judgement: Low / Moderate / Serious / Critical / No information.

Newcastle-Ottawa Scale (NOS)
Cohort studies: max 9 stars (Selection x4, Comparability x2, Outcome x3)
Case-control: max 9 stars (Selection x4, Comparability x2, Exposure x3)
Thresholds: 7-9 = Good (low risk) | 5-6 = Fair (moderate) | 0-4 = Poor (high risk)
AXIS (Cross-sectional)

20 items across 5 sections (Introduction, Methods, Results, Discussion, Other). Each Yes/No/Unclear. Section summary scores presented as ratios.

JBI Critical Appraisal (RCT)

13 items covering randomization, allocation concealment, blinding, completeness of follow-up, intention-to-treat analysis, outcome measurement, and statistics. Each Yes / No / Unclear / Not applicable.

References: Sterne JAC et al. BMJ. 2019;366:l4898. | Sterne JAC et al. BMJ. 2016;355:i4919. | Wells GA et al. Ottawa Hospital Research Institute. | Downes MJ et al. BMJ Open. 2016;6:e011458. | Barker TH et al. JBI Evid Synth. 2023;21(3):494-506.

Licensing: RoB 2, ROBINS-I and JBI are CC BY-NC-ND 4.0; AXIS is CC BY-NC 4.0; NOS has no formal license stated. Reproduced here with attribution for non-commercial research use. See each tool's page for the citation and license note.

GRADE Evidence Certainty

Starting certainty: RCT = 4 (High) | Observational = 2 (Low)

Downgrade (-1 or -2 each):
1. Risk of bias 2. Inconsistency 3. Indirectness 4. Imprecision 5. Publication bias

Upgrade (+1 each):
1. Large effect (RR >2 or <0.5) 2. Dose-response 3. Plausible confounders reduce effect

Final = max(1, min(4, start - downgrades + upgrades))
4=High 3=Moderate 2=Low 1=Very Low

Reference: Guyatt G et al. J Clin Epidemiol. 2011;64:383-394.

Randomization Algorithm

Seeded LCG (Linear Congruential Generator)
X(n+1) = (1664525 * X(n) + 1013904223) mod 2^32
Uniform [0,1] = X(n+1) / 2^32
Block Randomization
Block size b (fixed or random from user set)
Within each block: Fisher-Yates shuffle of group labels
Guarantees balance after every b participants

Reference: Schulz KF, Grimes DA. Lancet. 2002;359:515-519.

PRISMA 2020 & CONSORT Flow Diagrams

PRISMA 2020 Auto-calculations
Records screened = Total identified - Duplicates - Auto-removed
Reports sought = Screened - Excluded (title/abstract)
Reports assessed = Sought - Not retrieved
Included = Assessed - Excluded (full-text)
CONSORT 2010

Two-arm parallel RCT flow: enrollment, allocation, follow-up (lost/discontinued), analysis. All counts user-entered; SVG rendered with fixed coordinate layout.

References: Page MJ et al. BMJ. 2021;372:n71. | Schulz KF et al. BMJ. 2010;340:c332.

AI-Powered Features (Extraction, Screening, Advisor)

AI features use large language models (LLMs) via API. These are assistance tools, not autonomous decision-makers.

AI Data Extraction

Extracts structured fields from PDFs with source quotes and confidence ratings (High/Medium/Low). All output must be independently verified by a trained reviewer.

AI Screening

Scores references against user-defined inclusion/exclusion criteria. Produces include/exclude recommendation with reasoning. Final decisions must be made by human reviewers.

AI Data Advisor

Analyses uploaded datasets and recommends appropriate statistical methods. Generates R and Python code snippets. Users must verify appropriateness for their specific research question.

Figure Studio (Publication Figures)

Figure Studio renders publication figures in-browser from pasted or piped data, with control over labels, fonts, palette, height, gridlines, and a log x-axis for ratio measures. Each figure type ships a sample dataset and a downloadable template, and meta-analysis results can be sent straight to it.

Effect-modification figures

Two views for showing how an exposure effect varies across strata of a third variable. The interaction plot draws the outcome across levels of the modifier, one line per exposure group: non-parallel lines indicate effect modification. The subgroup forest plots the effect estimate and 95% CI within each stratum against a null reference line, with the overall/pooled row drawn as a diamond.

Forest and other charts

Forest plots size markers by weight and mark the pooled estimate as a diamond. Charts export as PNG and SVG. Rendering uses Plotly.js (MIT licensed).

References: VanderWeele TJ, Knol MJ. A tutorial on interaction. Epidemiol Methods. 2014;3(1):33-72. | Plotly Technologies Inc. Collaborative data science. Montréal, QC; 2015.

Support

Get help with your account, report a problem, or find answers to common questions.

Email support
We reply within 1–2 business days.
help@covexe.com
In-app feedback
Report a bug, suggest a feature, or send a screenshot directly.
Account & access
Do I need an account to use Kovex?
No. Every statistical tool works without signing in. Your work is saved in your browser. Signing in adds cloud sync (access your work from any device) and the shared AI features.
I forgot my password.
On the login screen, click Forgot password? and enter your email. You will receive a reset link within a few minutes. Check your spam folder if it does not arrive.
The AI feature says I need to sign in.
Shared AI requires a free account to prevent abuse. As an alternative, paste your own OpenAI, Anthropic, or Gemini key in API Key (sidebar) to use AI features without an account.
How do I delete my account and data?
Click your name in the top-right corner of the app and choose Delete account. You will be asked to confirm by typing DELETE. Deletion is immediate and permanent. If you have trouble signing in, email help@covexe.com and we will delete it manually within 30 days.
Data & privacy
Is my research data stored on your servers?
No. All statistical calculations run entirely in your browser and no study data is transmitted to our servers. If you sign in and save a project, the project contents are stored in our encrypted cloud database. See the Privacy Policy for full details.
What data does the AI feature send?
Only the text or PDF content you actively submit to an AI feature is sent, encrypted in transit, to a third-party language model provider (OpenAI, Anthropic, or Google). Kovex does not store this on its own servers. Do not submit patient-identifiable data.
I want to exercise a privacy right (access, correction, deletion).
Email help@covexe.com with the subject "Privacy Request". We respond within 30 days as required by GDPR and Taiwan PDPA. See the Legal & Privacy page for your full rights.
Bugs & feature requests
Something is not working correctly.
Use the Send Feedback button and choose Bug. A screenshot is very helpful. Include which tool you were using and what you expected to happen.
I have a feature idea or methodology request.
Use Send Feedback and choose New tool or Recommendation. We read every submission.
Looking for how-to guides?
The User Manual covers the full systematic review workflow, every tool, data formats, and step-by-step guidance.

Figure Studio

Pick a pre-built figure type, map your columns, and get the same clean, publication-ready design every time. The layout, fonts, colours, and spacing are locked to a house style, so the figure is reproducible and consistent, not improvised.

What this does

Produces consistent, reproducible figures from pre-built templates. Unlike asking an AI to "draw a chart" (which differs every time), each template renders with a fixed design, so the same data always yields the same figure.

When to use

Whenever you need a clean standard figure (grouped bars with error bars, box plots, scatter with fit, multi-series line, ranked bars, histogram) without fiddling with styling.

How to use

  • Paste data (first row = headers) or load the sample
  • Choose a figure type
  • Confirm the auto-mapped columns, or adjust them
  • Set title and labels, pick a colour palette
  • Export PNG or SVG

Interpreting results

Error bars show the column you map to "error" (for example a 95% CI half-width or SD). Scatter fit is ordinary least squares with the equation and R-squared shown.

Design follows common figure conventions; rendering is deterministic.
1. Figure type
Choose a figure. Its sample data loads automatically so you can see the layout, then replace it with your own.
2. Data
Paste tab or comma separated data. First row = column headers. Keep the same column shape as the sample.
3. Style
Title
X label
Y label
Palette
More styling
Font
Legend
Title size
Axis label size
Number size
Legend size
Height (px)
Width (px)
Line width
Marker size
X min
X max
Y min
Y max
Background
Annotation color
Gridlines
Log x-axis
Build with AI
Have a different figure in mind? Describe it. The AI picks the closest consistent template, invents matching sample data so you can see the layout, and builds it. Then replace the sample with your own data and click Generate figure. You will not lose detail or have to re-prompt: the design is locked, only what you ask changes.
Ideas to describe
Saved figures
Save a figure setup to reopen and tweak later, with all chart controls.
Figure
Paste data and choose a figure type.

Project Home

The home base for your active project. Every tool you use saves into this project automatically. Run blinded dual screening, resolve conflicts, and manage extraction templates here.

What this does

Your project workspace: a progress overview plus blinded dual screening, conflict resolution, and custom extraction templates. All tool data is auto-saved to the open project.

When to use

Throughout a review. Open a project, then work across the tools; come back here to screen references with two reviewers and resolve disagreements.

How to use

  • Open or create a project (Projects button, top right)
  • Use any tool and your inputs are saved to the project automatically
  • Import references and have each reviewer vote independently
  • Resolve conflicts and review inter-rater agreement (kappa)
  • Build extraction templates and send them to the Extraction tool

Interpreting results

Cohen's kappa: <0.20 slight, 0.21-0.40 fair, 0.41-0.60 moderate, 0.61-0.80 substantial, >0.80 almost perfect agreement.

Cohen J, Educ Psychol Meas 1960; Cochrane Handbook Ch. 4

User Manual

A practical guide to KOVΞX LABS, organised by the systematic-review workflow. Each tool also has a "?" button on its own page for quick reference.

All Citations

Every methodology implemented in Kovex is backed by peer-reviewed literature. Copy individual citations or all at once for your Methods section.

What this does

Lists every academic reference backing the methods used in Kovex. Copy-ready for your manuscript.

When to use

When writing your methods or reference section and need to cite the statistical methods used.

How to use

  • Browse citations by tool
  • Click to copy individual citations
  • Use "Copy All" for a complete reference list

Interpreting results

This is a reference page. Include relevant citations in your manuscript to document which methods and software were used.

All citations are listed on this page
Cite what you use
Each tool in Kovex has a “How to cite” button on its page. This page collects all citations in one place for convenience.
Copy All Citations
All references as a numbered list - ready to paste into your Methods section
How to Cite Kovex

A formal way to cite Kovex is on the way. We are preparing a methods paper that validates the platform's quantitative output against reference implementations; once it is published you will be able to cite it here.

Citation coming soon. The Kovex methods paper is in preparation; a citable reference will appear here when it is published. Please always cite the underlying statistical methods and source studies in your review.
Transparency note: Full formulas, algorithms, and downloadable R replication code for every computational tool are available on the Methods page. We encourage independent verification of all results.

My Plan

Your subscription, AI usage, and team. Use the preview switch to see what each plan looks like; once billing is live, your real plan drives this automatically.

PDF Library

Read your papers inside KOVEX, with no more juggling tabs. PDF, Word, text, Markdown and LaTeX are supported. Ask the paper questions, jump straight to extraction, and keep every reference for a project in one place. Files are stored privately in this browser.

API Key

AI features (data extraction, screening, the assistant, figures, and the data advisor) are free when you connect your own AI provider key. Your key is stored only in this browser and is sent with your AI requests. It is never saved on our servers.

Connect your AI provider

PLANS

What this does

Shows available subscription plans and their included features.

When to use

When deciding which plan fits your research needs.

How to use

  • Compare plan features
  • Select the plan that matches your usage

Interpreting results

Free tier includes all core tools. Paid plans add AI-powered features and higher usage limits.

Monthly Yearly
Free
$0 /forever
Every tool, for everyone
All tools: full review pipeline, meta-analysis, NMA, 40+ statistics
PDF reader & study library: up to 3 papers
Unlimited AI with your own API key (OpenAI, Claude, Gemini)
10 managed AI extractions per month, no key needed
Cloud sync across devices with a free account
Citation import and export (RIS, BibTeX, EndNote)
PNG, SVG, CSV and Word report exports
Pro Popular
$90 /year
Save $18/yr
Everything in Free
PDF reader & study library: up to 100 papers
Ask-the-paper AI, highlights and reading tools
200 managed AI extractions per month, no key needed
AI screening, extraction and data advisor
Top up anytime with pay-as-you-go AI credits
Or connect your own key for unlimited AI
Priority AI processing
Verified student and academic discount
Team
$120 /year
Save $24/seat/yr
Per seat. Minimum 3 seats.
Everything in Pro, for every member
Shared PDF library: up to 1,000 papers
Shared cloud projects with real-time collaboration
Blinded dual screening, conflict resolution and Cohen's kappa
Roles and permissions (admin, reviewer, viewer)
Assign reviewers and split the workload
Team progress dashboard: per-member throughput and pending conflicts
Comments and @mentions on references
Shared reference library and reusable project templates
Each member can use their own AI key, unlimited and free
Optional managed AI add-on for the team
Word and Excel export and one-click reports
Email support
Institution
Custom
Universities and enterprises
Contact sales
Everything in Team, unlimited reviews and projects
Unlimited PDF library across the organization
Org-wide admin console: every review, project and member
Usage analytics and exportable reports for librarians and PIs
Single sign-on (SAML / OIDC) and SCIM user provisioning
Volume seats with central seat management
Full audit trail: 100% traceable changes for compliance
Pooled managed AI, or bring your org key (Vertex, Azure, OpenAI)
Data processing agreement, GDPR aligned
Custom onboarding, training and priority support (SLA)
Bring your own AI key on any plan (OpenAI, Anthropic / Claude, or Google / Gemini): AI is unlimited and free, and your key stays in your browser. Prefer not to manage a key? Paid plans include a monthly managed-AI allowance, with credits to top up, plus team collaboration.
Students and academics: a verified academic discount applies to Pro and Team. Contact us to verify your status.
FAQ
Is Kovex really free?

Yes. Every research tool is free for everyone, with no limits. AI features are free too when you connect your own API key.

What do I actually pay for?

Three optional things: managed AI (we run the AI so you do not need your own key), team collaboration with shared cloud projects, and institutional licenses.

Which AI providers can I use?

OpenAI, Anthropic (Claude), or Google (Gemini). Add your key on the AI Key page; it stays in your browser and is never stored on our servers.

Do you offer credits or academic deals?

Yes. Pay-as-you-go AI credits for occasional use, and discounted institutional licensing for universities. Contact us.