The page is taking longer to load.
This guide explains the complete AccuSynthesis workflow—from defining the project protocol and designing a data-extraction template to understanding dimensions, processing papers, reviewing results, and exporting evidence.
AccuSynthesis is a structured data-extraction workflow for systematic literature reviews. You define what information should be collected, upload the included publications, review the extraction design, and let the system produce evidence tables that remain editable and exportable.
It supports clinical effectiveness, safety, health economics, diagnostic accuracy, epidemiology, qualitative evidence, scoping reviews, and mixed evidence projects. The review type provides guidance; it does not force every project into the same template.
| What you provide | What AccuSynthesis prepares | What you receive |
|---|---|---|
| Project protocol and extraction goal | A reviewable extraction structure | Structured rows and columns |
| Research papers and optional supplements | Paper-specific row identities | Evidence linked to each paper |
| Manual, imported, converted, or generated template | Descriptions, groups, types, and constraints | Excel and Word exports |
Each stage remains reviewable. AI-generated content is presented as a draft before it becomes part of the project.
Give the project a clear title and provide a concise protocol describing the evidence scope.
Add the included papers and any supported supplementary documents, then confirm that text extraction is ready.
Build manually, import an AccuSynthesis workbook, convert a client workbook, or generate a draft from the protocol and goal.
Check sheets, sections, groups, columns, descriptions, shared dimensions, value types, and allowed options.
AccuSynthesis discovers relevant row identities and fills the requested fields while preserving template boundaries.
Inspect the extracted evidence, add more files when needed, and export all or selected results to Excel or Word.
The final project protocol is the main scope boundary used when generating descriptions and extracting data. It should describe the review—not the application or the extraction technology. The final normalized protocol can contain up to approximately 700 words.
Every method produces the same supported hierarchy, so the resulting template can be reviewed and edited in the builder.
| Method | Best used when | Required input | How AI is used | Review |
|---|---|---|---|---|
| Build manually | Small or highly customized forms | Protocol and user-defined fields | Optional description generation | Edit directly in builder |
| Import structured template | Reusing an exported AccuSynthesis template | Structured workbook | No reinterpretation required | Verify exact import |
| Convert raw workbook | Migrating a client extraction workbook | Protocol and raw workbook | Interprets headers and generates descriptions | Review converted draft |
| Generate with AI | Starting from review requirements | Protocol and goal; workbook/PDF summaries optional | Proposes the complete template | Review generated draft |
The hierarchy gives every short column name a complete meaning. A column named N is ambiguous by itself, but its full path can mean Outcomes → Efficacy → Overall Survival → N.
| Level | Purpose | Example |
|---|---|---|
| Sheet | A table with one row structure and a related subject. | Outcomes |
| Section | A broad division spanning related column groups. | Efficacy |
| Column group | Fields that must be interpreted together. | Overall Survival |
| Column | One value to extract, with a stable key and description. | Hazard ratio |
This is the most important template-design decision. A dimension identifies what a row represents. A measure describes the result represented by that row.
| Row mode | Use it when | Example |
|---|---|---|
| One row per article | The requested values normally occur once for the paper or study. | Citation details, study design, country |
| Multiple rows per article | Values vary across reported arms, populations, outcomes, events, or timepoints. | Clinical outcomes, adverse events, economic scenarios |
Multi-row sheets expose their shared dimensions before extraction. You can add, remove, reorder, rename, enable, or disable them. Existing clinical projects retain Treatment Name, Analysis Group, Subgroup Type, and Timepoint as their default structure, but other evidence domains can use different dimensions.
| Review or sheet type | Common dimensions | Common measures or values |
|---|---|---|
| Clinical effectiveness | Treatment, analysis population, subgroup, outcome, timepoint | N, events, mean, SD, HR/RR/OR, CI, p-value |
| Safety / adverse events | Treatment, safety population, event name, grade, timepoint | Evaluated N, event N, percentage, relative effect |
| Health economics | Intervention, comparator, population, scenario, outcome, time horizon | Cost, QALYs, incremental values, ICER, utility |
| Diagnostic accuracy | Index test, threshold, population, reference standard | TP, FP, TN, FN, sensitivity, specificity, AUC |
| Epidemiology | Population, exposure, outcome, region, age/sex group, period | Sample N, cases, prevalence, incidence, RR/OR/HR |
| Qualitative evidence | Theme, subtheme, participant group, setting, context | Finding, quotation, interpretation, confidence |
A column can be labelled as a dimension candidate in Column settings, but that label alone does not change how rows are created. To make its reported values identify separate rows, move the column into Row Structure as either a shared or local row dimension.
The builder uses the wording Use as shared row dimension and Use as local row dimension. This action moves the field—it does not copy it—so the same name is not sent twice to AI or shown twice in the result and exported workbook.
| Choice | Use it when | Where it applies | Example |
|---|---|---|---|
| Shared row dimension | The field identifies rows across most or all groups in the sheet. | The sheet and its row schemas. | Treatment, analysis population, subgroup, timepoint |
| Local row dimension | The field identifies rows only for one repeating type of result. | The selected row schema and every group sharing that schema. | Outcome Name, Instrument, Event Name, Grade, Economic Scenario |
| Ordinary measure column | The field describes a value for an existing row and does not decide whether another row is needed. | Its own column group. | N, mean, SD, HR, CI, event percentage, ICER |
| Concept | What it means | When to use it |
|---|---|---|
| Fixed group | Uses the sheet's shared dimensions and introduces no local dimensions of its own. | The columns have a known meaning in the header, such as dedicated OS HR and PFS HR groups. |
| Repeating group | Uses a row schema with one or more local dimensions. | Names vary by paper, such as Outcome Name, Event Name, Grade, Theme, or Scenario. |
| Row schema | The reusable definition of the shared and local values that together identify one row. | A repeating group needs a clear rule for what makes two rows different. |
| Groups sharing one row schema | Two or more repeating groups use exactly the same canonical row list and are merged side by side. | Only when all those groups must align to the same dimension combinations. |
| Independent row schemas | Groups discover separate row lists and are appended as different result-row types. | Outcome rows and adverse-event rows have different identities or granularity. |
Suppose a repeating Outcome Result group initially contains these ordinary columns:
| Before | Role in the intended table | Builder action |
|---|---|---|
| Outcome Name | Determines whether the row is OS, PFS, response rate, or another reported outcome. | Set to Row dimension, then choose Use as local row dimension. |
| N | A value reported for the outcome row. | Keep as an extracted value. |
| Estimate | A value reported for the outcome row. | Keep as an extracted value. |
| Confidence Interval | A value reported for the outcome row. | Keep as an extracted value. |
After the move, the structure becomes:
| Row schema | Local dimension | Measure columns | Possible reported rows |
|---|---|---|---|
| Outcome Result rows | Outcome Name | N, Estimate, Confidence Interval | Overall Survival; Progression-Free Survival |
Overall survival and progression-free survival have their own measure columns. Outcome identity is encoded in the column paths, so it does not create additional rows.
| Study | Treatment | Timepoint | OS HR | OS CI | PFS HR | PFS CI |
|---|---|---|---|---|---|---|
| Study 1 | Drug A | 12 months | 0.72 | 0.58–0.89 | 0.81 | 0.67–0.98 |
| Study 1 | Control | 12 months | Reference | NR | Reference | NR |
Design lesson: Use this pattern when the outcome set is known and stable. It creates a wider table with fewer rows.
A generic Outcome Name dimension identifies each reported outcome, while the same measure columns are reused.
| Study | Treatment | Outcome Name | Timepoint | N | Estimate | CI |
|---|---|---|---|---|---|---|
| Study 1 | Drug A | Overall survival | 12 months | 240 | 0.72 | 0.58–0.89 |
| Study 1 | Drug A | Progression-free survival | 12 months | 240 | 0.81 | 0.67–0.98 |
Design lesson: Use this pattern when outcome names vary or are not known in advance. It creates a narrower table with more rows.
Event name and grade identify distinct reported rows. Measures describe how often each event occurred.
| Study | Treatment | Event Name | Grade | Evaluated N | Event N | Percentage |
|---|---|---|---|---|---|---|
| Study 1 | Drug A | Nausea | Any grade | 150 | 42 | 28.0 |
| Study 1 | Drug A | Nausea | Grade ≥3 | 150 | 6 | 4.0 |
| Study 1 | Drug A | Fatigue | Any grade | 150 | 30 | 20.0 |
Design lesson: Create only event and grade combinations actually reported by the paper.
Scenario and time horizon can identify distinct economic results even when the intervention and comparator remain unchanged.
| Study | Intervention | Comparator | Scenario | Time Horizon | Incremental Cost | Incremental QALY | ICER |
|---|---|---|---|---|---|---|---|
| Study 2 | Intervention A | Usual care | Base case | Lifetime | 12,500 | 0.62 | 20,161/QALY |
| Study 2 | Intervention A | Usual care | Sensitivity analysis | 10 years | 9,800 | 0.41 | 23,902/QALY |
Design lesson: Dimensions are domain-neutral. They should describe whatever makes one reported result different from another.
Use the tune control beside a column to define the expected content. Clear constraints improve consistency and make exported templates easier to use.
| Setting | Choices | Meaning | Example |
|---|---|---|---|
| Role | Article-level, dimension, measure | Describes semantic purpose. A dimension candidate becomes active only after it is moved into Row Structure. | Outcome Name can be moved to a local dimension; hazard ratio remains a measure. |
| Value type | Text, number, select one, select multiple, yes/no, date | Describes the form of the expected value. | Randomized is Yes/No. |
| Allowed options | User-defined list | Preferred or closed vocabulary for select fields. | RCT, cohort, case-control |
| Discover values | On or off | Controls whether a clearly reported unlisted value may be retained. | Allow a previously unseen country. |
The extracted-data viewer follows the same hierarchy and color logic as the Excel-layout preview, making groups easier to recognize before export.
| Capability | What it provides |
|---|---|
| Extracted-data viewer | Scrollable sheet tables with row dimensions, sections, groups, and columns. |
| Data-availability status | Shows Extracted, No eligible data, Partial, or Failed for each paper, worksheet, and row schema. No eligible data includes a reason but does not create an artificial all-NR result row. |
| Excel export | Colored hierarchical headers, filters, configured dimensions, extracted rows, and a separate Extraction Status audit sheet. |
| Word export | Structured tables retaining the template hierarchy, selected evidence, and extraction-status audit information. |
| Selective export | Export all results or choose specific sheets and column groups. |
| Reusable template | Protocol, descriptions, roles, types, options, dimensions, and hidden structured metadata for exact re-import. |
| Do | Avoid |
|---|---|
| check_circle Use one clear concept per column. | cancel Combine unrelated values in a single field. |
| check_circle Keep related statistics in the same semantic group. | cancel Mix survival, safety, and quality-of-life measures without hierarchy. |
| check_circle Use dimensions only when a changed value identifies another result. | cancel Turn every descriptive field into a dimension. |
| check_circle Write descriptions that state evidence, unit, and expected format. | cancel Repeat only the column name in the description. |
| check_circle Use separate sheets when evidence families have different row structures. | cancel Create one extremely sparse sheet for unrelated evidence. |
| check_circle Use closed options for stable vocabularies. | cancel Use closed lists when legitimate values are unpredictable. |
