study11 fields
| study_design | RCTNon-randomized TrialProspective CohortRetrospective CohortCase-ControlCross-sectionalOther | Study design classification. RCT requires TRUE random allocation (computer-generated sequence, random-number table, sealed random envelopes). A controlled comparison where the investigator assigned the intervention by a systematic non-random rule — alternation, enrolment date/month, record number, odd/even — is a Non-randomized Trial, NOT a cohort. Do not classify RCT on the word 'randomized' alone; check the described allocation method. Cohort/Case-Control/Cross-sectional apply only to observational studies where exposure is not investigator-assigned; for those, forward data collection = Prospective Cohort, existing records = Retrospective Cohort. |
| country | text | Country of patient enrollment. MUST use ISO 3166-1 short names, except: always write 'UK' (not 'United Kingdom') and 'USA' (not 'United States of America' or 'United States'). Multiple countries comma-separated, alphabetical. Use country of enrollment, not author affiliation (e.g., 'France, Spain, UK, USA'). |
| center_type | singlemulti | Single-center or multicenter study. Multicenter requires patients enrolled at >=2 distinct sites. National registry studies = Multicenter. |
| enrollment_criteria | textharmonize | Key inclusion AND exclusion criteria as reported. Limit to 5-7 criteria: age, condition, procedure, severity, exclusions (prefix with 'no'). Use parentheses to group alternatives with 'or' (e.g., 'age >=18 & (hip or knee) arthroplasty & ASA I-III & no revision surgery'). |
| enrollment_period | textharmonize | Date range of patient enrollment. Format: 'YYYY-MM to YYYY-MM' or 'YYYY to YYYY' if months not reported. Use enrollment dates, not publication or data collection dates (e.g., '2015-01 to 2020-12'). |
| funding_source | IndustryNon-industryMixedDeclared none | Primary funding source AND/OR conflict-of-interest disclosure. Industry = any pharmaceutical/device/commercial company funded the work. Non-industry = government/academic/foundation/charity only. Mixed = both. Declared none = positive disclosure of no funding or no COI with no offsetting positive funding signal. When a positive funding signal is present, use that; only fall through to Declared none if no funding source is named. |
| sample_size_total | number | Total participants allocated/enrolled across the on-axis arms — the sum of the per-unit sample_size (the pre-attrition denominator), matching that field's stage. Not the screened, eligible, or source-population/registry denominator, and not the analyzed/complete-case total. If not explicitly stated, sum the per-unit sample_size values. |
| data_source | textharmonize | Primary data source for the analysis. Format as a short noun phrase that names the source type and, when specified, the named registry or database (e.g., 'institutional database', 'national registry', 'UNOS/OPTN registry', 'NHANES', 'claims database', 'prospective single-center cohort'). When the paper draws from multiple sources, comma-separate. Use the source named in Methods, not author affiliation. |
| surgery_type | textharmonize | Type(s) of surgery in the study population as a short noun phrase drawn from the paper's population description (e.g., "cardiac surgery", "elective orthopedic", "major abdominal"). When multiple distinct surgery classes are enrolled, comma-separate the principal classes. Use the operative procedure category, not the underlying disease alone. |
| anesthetic_type | generalneuraxialperipheral_regionalcombination | Predominant anesthetic modality for the index procedure. general = general anesthesia alone; neuraxial = spinal or epidural (with or without sedation) and no general; peripheral_regional = peripheral nerve block without neuraxial or general as the primary modality; combination = more than one of the above modalities used across the cohort or within patients (e.g., general plus epidural). Keep neuraxial and peripheral_regional distinct — do not merge under a single 'regional' label. Classify from the methods' anesthetic description; not_reported when the paper is silent on anesthetic modality. |
| preop_cognitive_assessment | text | Whether a baseline cognitive assessment was performed before the cognitive intervention, its timing (strictly preoperative vs post-surgery-but-pre-intervention), and whether patients with preexisting cognitive impairment were identified, graded, or excluded. Format as a short phrase covering both assessment/timing and impairment handling (e.g., "MoCA preoperatively, impairment excluded"; "baseline battery post-CABG pre-intervention, impairment not graded"; "no preoperative assessment"). not_reported only when the paper is silent on BOTH a baseline assessment and impairment handling. |
unit14 fields
| sample_size | number | Participants allocated/enrolled to this unit — the randomized or baseline-characteristics-table N, before attrition. Do NOT use the analyzed/complete-case N when they differ under dropout. Prefer the baseline Table 1 N; fall back to the CONSORT 'allocated' count. |
| age_years | average | Age at enrollment/baseline in years. |
| female_percent | number | Percentage female in this unit. Calculate the complement if only male is given (60% male = 40% female). If the cohort is sex-restricted by procedure or eligibility, infer accordingly (a female-only procedure → 100; a male-only procedure → 0). |
| withdrawals | number | Participants in this unit who withdrew consent or discontinued participation between randomization/enrollment and final analysis. Count only consent-withdrawal and active discontinuation — do not include participants who remained enrolled but were not reached for follow-up. |
| lost_to_follow_up | number | Participants in this unit who remained enrolled but could not be reached for the final outcome assessment. Count only contact-loss with retained consent. If the paper reports a single combined count without distinguishing consent-withdrawal from contact-loss, record the combined count here and note the combined nature in the comment. |
| intervention_type | textharmonize | The cognitive training or perioperative care this unit received, as a short noun phrase naming the delivered activity (e.g., "tablet-based cognitive exercises", "usual care", "sham non-adaptive puzzles", "physical prehabilitation"). Capture the paper's label for what was administered to this arm, not the review arm label. |
| training_sessions_total | number | Total number of cognitive-training sessions prescribed for this unit (count of planned sessions across the full program). not_applicable for arms with no cognitive training (standard_care or sham_control without a training schedule). Prefer the protocol-prescribed total over sessions actually completed; completed-session counts belong in training_adherence. |
| training_frequency | text | How often cognitive-training sessions were scheduled, as a short phrase (e.g., "daily", "3x/week", "twice daily"). not_applicable for arms with no cognitive training. Report the planned schedule, not adherence-adjusted frequency. |
| session_duration_minutes | number | Planned duration of each cognitive-training session in minutes. Convert from hours if reported that way (× 60). Use the per-session planned length, not total program contact time. not_applicable for arms with no cognitive training. |
| training_overall_duration_days | number | Overall length of the cognitive-training program in days, from first to last scheduled session. Convert from weeks (× 7) or months (× 30) so values remain whole and comparable; do not leave in mixed units. not_applicable for arms with no cognitive training. |
| training_delivery | textharmonize | How and by whom the cognitive training was administered, as a short phrase covering modality and supervisor (e.g., "self-administered app", "therapist-led in person", "neuropsychologist-supervised", "researcher-supervised tablet"). not_applicable for arms with no cognitive training. |
| training_timing | preoperativepostoperativeperioperative | When this unit's cognitive training was delivered relative to the index surgery. preoperative = entirely before surgery; postoperative = entirely after surgery; perioperative = spanning both pre- and post-operative periods. Required for the pre-/post-/peri-operative subgroup analysis — classify from the intervention schedule, not from when outcomes were assessed. not_applicable for an arm receiving no cognitive training. |
| training_adherence | number | Proportion of prescribed cognitive-training sessions this unit actually completed, as a percentage. Use the paper's own adherence, compliance, or completion figure directly; do not derive a rate from dropout or withdrawal counts alone. not_applicable for an arm with no cognitive training; not_reported when the paper gives no completion/adherence figure. |
| preop_cognitive_impairment_percent | number | Percentage of this unit with preexisting cognitive impairment at baseline, per the paper's stated definition or threshold. Calculate complement if only the unimpaired proportion is given. not_reported if the paper does not state the proportion in this unit (including when impairment was an exclusion and no baseline prevalence is given). |
outcomes6 outcomes · 33 fields
| delayed_neurocognitive_recovery5 | ||
| delayed_neurocognitive_recovery_rate | rate | Rate of delayed_neurocognitive_recovery in this unit (events / evaluated / percent). Include only confirmed cases; exclude unconfirmed or suspected cases unless the paper explicitly counts them. |
| delayed_neurocognitive_recovery_timepoint | text | Time at which delayed_neurocognitive_recovery was assessed in this unit. Format: relative-to-baseline duration or named clinical milestone (e.g., '30 days post-randomization', '1 year', 'in-hospital', 'discharge'). When multiple timepoints are reported, use the protocol's primary or pre-specified one. |
| delayed_neurocognitive_recovery_definition_as_reported | textharmonize | Author's stated definition, grading criteria, threshold, or measurement modality for delayed_neurocognitive_recovery. Use the canonical instrument name when applicable (the named scale, grading system, or clinical-criteria set the paper cites); otherwise a concise paraphrase capturing the paper's diagnostic criterion or definitional phrase ('was defined as', 'diagnosed by', 'graded using'). Keep to a short noun phrase or single clause. |
| delayed_neurocognitive_recovery_effect_vs_standard_care | effect_estimate | Adjusted effect estimate for delayed_neurocognitive_recovery, comparing the extraction unit (the arm named in the '## Extraction unit' block) against the 'standard_care' reference arm (see the '## Comparison axis' block), over the whole study follow-up — one estimate per outcome, NOT a per-timepoint value. Populate on any unit other than 'standard_care' itself, which is not_applicable. If the paper reports no delayed_neurocognitive_recovery effect for this arm vs 'standard_care', mark not_reported. Give measure and adjustment per the effect_estimate shape; note the adjustment model/covariates and follow-up horizon in the comment. |
| delayed_neurocognitive_recovery_effect_vs_sham_control | effect_estimate | Adjusted effect estimate for delayed_neurocognitive_recovery, comparing the extraction unit (the arm named in the '## Extraction unit' block) against the 'sham_control' reference arm (see the '## Comparison axis' block), over the whole study follow-up — one estimate per outcome, NOT a per-timepoint value. Populate on any unit other than 'sham_control' itself, which is not_applicable. If the paper reports no delayed_neurocognitive_recovery effect for this arm vs 'sham_control', mark not_reported. Give measure and adjustment per the effect_estimate shape; note the adjustment model/covariates and follow-up horizon in the comment. |
| postoperative_neurocognitive_disorder5 | ||
| postoperative_neurocognitive_disorder_rate | rate | Rate of postoperative_neurocognitive_disorder in this unit (events / evaluated / percent). Include only confirmed cases; exclude unconfirmed or suspected cases unless the paper explicitly counts them. |
| postoperative_neurocognitive_disorder_timepoint | text | Time at which postoperative_neurocognitive_disorder was assessed in this unit. Format: relative-to-baseline duration or named clinical milestone (e.g., '30 days post-randomization', '1 year', 'in-hospital', 'discharge'). When multiple timepoints are reported, use the protocol's primary or pre-specified one. |
| postoperative_neurocognitive_disorder_definition_as_reported | textharmonize | Author's stated definition, grading criteria, threshold, or measurement modality for postoperative_neurocognitive_disorder. Use the canonical instrument name when applicable (the named scale, grading system, or clinical-criteria set the paper cites); otherwise a concise paraphrase capturing the paper's diagnostic criterion or definitional phrase ('was defined as', 'diagnosed by', 'graded using'). Keep to a short noun phrase or single clause. |
| postoperative_neurocognitive_disorder_effect_vs_standard_care | effect_estimate | Adjusted effect estimate for postoperative_neurocognitive_disorder, comparing the extraction unit (the arm named in the '## Extraction unit' block) against the 'standard_care' reference arm (see the '## Comparison axis' block), over the whole study follow-up — one estimate per outcome, NOT a per-timepoint value. Populate on any unit other than 'standard_care' itself, which is not_applicable. If the paper reports no postoperative_neurocognitive_disorder effect for this arm vs 'standard_care', mark not_reported. Give measure and adjustment per the effect_estimate shape; note the adjustment model/covariates and follow-up horizon in the comment. |
| postoperative_neurocognitive_disorder_effect_vs_sham_control | effect_estimate | Adjusted effect estimate for postoperative_neurocognitive_disorder, comparing the extraction unit (the arm named in the '## Extraction unit' block) against the 'sham_control' reference arm (see the '## Comparison axis' block), over the whole study follow-up — one estimate per outcome, NOT a per-timepoint value. Populate on any unit other than 'sham_control' itself, which is not_applicable. If the paper reports no postoperative_neurocognitive_disorder effect for this arm vs 'sham_control', mark not_reported. Give measure and adjustment per the effect_estimate shape; note the adjustment model/covariates and follow-up horizon in the comment. |
| postoperative_delirium5 | ||
| postoperative_delirium_rate | rate | Rate of postoperative_delirium in this unit (events / evaluated / percent). Include only confirmed cases; exclude unconfirmed or suspected cases unless the paper explicitly counts them. |
| postoperative_delirium_timepoint | text | Time at which postoperative_delirium was assessed in this unit. Format: relative-to-baseline duration or named clinical milestone (e.g., '30 days post-randomization', '1 year', 'in-hospital', 'discharge'). When multiple timepoints are reported, use the protocol's primary or pre-specified one. |
| postoperative_delirium_definition_as_reported | textharmonize | Author's stated definition, grading criteria, threshold, or measurement modality for postoperative_delirium. Use the canonical instrument name when applicable (the named scale, grading system, or clinical-criteria set the paper cites); otherwise a concise paraphrase capturing the paper's diagnostic criterion or definitional phrase ('was defined as', 'diagnosed by', 'graded using'). Keep to a short noun phrase or single clause. |
| postoperative_delirium_effect_vs_standard_care | effect_estimate | Adjusted effect estimate for postoperative_delirium, comparing the extraction unit (the arm named in the '## Extraction unit' block) against the 'standard_care' reference arm (see the '## Comparison axis' block), over the whole study follow-up — one estimate per outcome, NOT a per-timepoint value. Populate on any unit other than 'standard_care' itself, which is not_applicable. If the paper reports no postoperative_delirium effect for this arm vs 'standard_care', mark not_reported. Give measure and adjustment per the effect_estimate shape; note the adjustment model/covariates and follow-up horizon in the comment. |
| postoperative_delirium_effect_vs_sham_control | effect_estimate | Adjusted effect estimate for postoperative_delirium, comparing the extraction unit (the arm named in the '## Extraction unit' block) against the 'sham_control' reference arm (see the '## Comparison axis' block), over the whole study follow-up — one estimate per outcome, NOT a per-timepoint value. Populate on any unit other than 'sham_control' itself, which is not_applicable. If the paper reports no postoperative_delirium effect for this arm vs 'sham_control', mark not_reported. Give measure and adjustment per the effect_estimate shape; note the adjustment model/covariates and follow-up horizon in the comment. |
| executive_function6 | ||
| executive_function | average | Value of executive_function in this unit, in the units the paper reports. Record per the average field shape: mean+SD, median+IQR, or median+min/max — include whichever the paper reports. Convert units when the field name specifies one; record verbatim otherwise. |
| executive_function_timepoint | text | Time at which executive_function was measured in this unit. Format: relative-to-baseline duration or named clinical milestone (e.g., 'baseline', 'day 1', 'discharge'). When measured at multiple timepoints, use the protocol's primary or pre-specified one. |
| executive_function_definition_as_reported | textharmonize | Author's stated measurement method for executive_function. Use the canonical instrument/assay/scale name when applicable (the named measurement device, lab assay, or rating scale the paper cites); otherwise a concise paraphrase capturing what was measured and how ('measured by', 'assessed with', 'derived from'). Keep to a short noun phrase or single clause. |
| executive_function_effect_vs_standard_care | effect_estimate | Adjusted effect estimate for executive_function, comparing the extraction unit (the arm named in the '## Extraction unit' block) against the 'standard_care' reference arm (see the '## Comparison axis' block), over the whole study follow-up — one estimate per outcome, NOT a per-timepoint value. Populate on any unit other than 'standard_care' itself, which is not_applicable. If the paper reports no executive_function effect for this arm vs 'standard_care', mark not_reported. Give measure and adjustment per the effect_estimate shape; note the adjustment model/covariates and follow-up horizon in the comment. |
| executive_function_effect_vs_sham_control | effect_estimate | Adjusted effect estimate for executive_function, comparing the extraction unit (the arm named in the '## Extraction unit' block) against the 'sham_control' reference arm (see the '## Comparison axis' block), over the whole study follow-up — one estimate per outcome, NOT a per-timepoint value. Populate on any unit other than 'sham_control' itself, which is not_applicable. If the paper reports no executive_function effect for this arm vs 'sham_control', mark not_reported. Give measure and adjustment per the effect_estimate shape; note the adjustment model/covariates and follow-up horizon in the comment. |
| executive_function_direction | higher_betterhigher_worse | Whether a higher score on the study's reported executive-function instrument indicates better cognition (higher_better: e.g., category fluency count, correct trials) or worse cognition (higher_worse: e.g., Trail Making Test completion time, error counts, Stroop interference time). Infer from the named instrument; not_reported only when the instrument is unnamed or non-standard — do not mark not_reported merely because the paper does not restate score direction. |
| global_cognitive_function6 | ||
| global_cognitive_function | average | Value of global_cognitive_function in this unit, in the units the paper reports. Record per the average field shape: mean+SD, median+IQR, or median+min/max — include whichever the paper reports. Convert units when the field name specifies one; record verbatim otherwise. |
| global_cognitive_function_timepoint | text | Time at which global_cognitive_function was measured in this unit. Format: relative-to-baseline duration or named clinical milestone (e.g., 'baseline', 'day 1', 'discharge'). When measured at multiple timepoints, use the protocol's primary or pre-specified one. |
| global_cognitive_function_definition_as_reported | textharmonize | Author's stated measurement method for global_cognitive_function. Use the canonical instrument/assay/scale name when applicable (the named measurement device, lab assay, or rating scale the paper cites); otherwise a concise paraphrase capturing what was measured and how ('measured by', 'assessed with', 'derived from'). Keep to a short noun phrase or single clause. |
| global_cognitive_function_effect_vs_standard_care | effect_estimate | Adjusted effect estimate for global_cognitive_function, comparing the extraction unit (the arm named in the '## Extraction unit' block) against the 'standard_care' reference arm (see the '## Comparison axis' block), over the whole study follow-up — one estimate per outcome, NOT a per-timepoint value. Populate on any unit other than 'standard_care' itself, which is not_applicable. If the paper reports no global_cognitive_function effect for this arm vs 'standard_care', mark not_reported. Give measure and adjustment per the effect_estimate shape; note the adjustment model/covariates and follow-up horizon in the comment. |
| global_cognitive_function_effect_vs_sham_control | effect_estimate | Adjusted effect estimate for global_cognitive_function, comparing the extraction unit (the arm named in the '## Extraction unit' block) against the 'sham_control' reference arm (see the '## Comparison axis' block), over the whole study follow-up — one estimate per outcome, NOT a per-timepoint value. Populate on any unit other than 'sham_control' itself, which is not_applicable. If the paper reports no global_cognitive_function effect for this arm vs 'sham_control', mark not_reported. Give measure and adjustment per the effect_estimate shape; note the adjustment model/covariates and follow-up horizon in the comment. |
| global_cognitive_function_direction | higher_betterhigher_worse | Whether a higher score on the study's reported global cognitive instrument indicates better cognition (higher_better: MoCA, MMSE, most composite neuropsychological z-scores) or worse cognition (higher_worse: instruments where larger values reflect greater deficit). Infer from the named instrument; not_reported only when the instrument is unnamed or non-standard — do not mark not_reported merely because the paper does not restate score direction. |
| cognitive_test_performance6 | ||
| cognitive_test_performance | average | Value of cognitive_test_performance in this unit, in the units the paper reports. Record per the average field shape: mean+SD, median+IQR, or median+min/max — include whichever the paper reports. Convert units when the field name specifies one; record verbatim otherwise. |
| cognitive_test_performance_timepoint | text | Time at which cognitive_test_performance was measured in this unit. Format: relative-to-baseline duration or named clinical milestone (e.g., 'baseline', 'day 1', 'discharge'). When measured at multiple timepoints, use the protocol's primary or pre-specified one. |
| cognitive_test_performance_definition_as_reported | textharmonize | Author's stated measurement method for cognitive_test_performance. Use the canonical instrument/assay/scale name when applicable (the named measurement device, lab assay, or rating scale the paper cites); otherwise a concise paraphrase capturing what was measured and how ('measured by', 'assessed with', 'derived from'). Keep to a short noun phrase or single clause. |
| cognitive_test_performance_effect_vs_standard_care | effect_estimate | Adjusted effect estimate for cognitive_test_performance, comparing the extraction unit (the arm named in the '## Extraction unit' block) against the 'standard_care' reference arm (see the '## Comparison axis' block), over the whole study follow-up — one estimate per outcome, NOT a per-timepoint value. Populate on any unit other than 'standard_care' itself, which is not_applicable. If the paper reports no cognitive_test_performance effect for this arm vs 'standard_care', mark not_reported. Give measure and adjustment per the effect_estimate shape; note the adjustment model/covariates and follow-up horizon in the comment. |
| cognitive_test_performance_effect_vs_sham_control | effect_estimate | Adjusted effect estimate for cognitive_test_performance, comparing the extraction unit (the arm named in the '## Extraction unit' block) against the 'sham_control' reference arm (see the '## Comparison axis' block), over the whole study follow-up — one estimate per outcome, NOT a per-timepoint value. Populate on any unit other than 'sham_control' itself, which is not_applicable. If the paper reports no cognitive_test_performance effect for this arm vs 'sham_control', mark not_reported. Give measure and adjustment per the effect_estimate shape; note the adjustment model/covariates and follow-up horizon in the comment. |
| cognitive_test_performance_direction | higher_betterhigher_worse | Whether a higher score on the study's reported domain-specific cognitive test (memory, attention, or processing speed — outside executive function) indicates better cognition (higher_better: e.g., words recalled, digit-span length) or worse cognition (higher_worse: e.g., reaction time, error counts). Infer from the named instrument; not_reported only when the instrument is unnamed or non-standard — do not mark not_reported merely because the paper does not restate score direction. |
rob14 fields
| rob2_randomization | LowSome concernsHigh | RoB 2 Domain 1: bias arising from the randomization process. LOW: random sequence (computer-generated, random number table) AND allocation adequately concealed (central randomization, sequentially numbered sealed opaque envelopes) AND baseline characteristics balanced. SOME CONCERNS: sequence generation or concealment not described, OR small baseline imbalances of uncertain origin. HIGH: predictable allocation (alternation, date of birth, day of week), OR substantial baseline imbalances suggesting failed randomization. |
| rob2_deviations | LowSome concernsHigh | RoB 2 Domain 2: bias due to deviations from intended interventions (effect of assignment). LOW: participants and personnel blinded OR no deviations from intended intervention arose OR appropriate ITT/mITT analysis estimating effect of assignment. SOME CONCERNS: minor deviations with unclear impact, OR analysis approach unclear. HIGH: substantial unbalanced deviations affecting the outcome, OR analysis was naive per-protocol/as-treated rather than ITT. |
| rob2_missing_data | LowSome concernsHigh | RoB 2 Domain 3: bias due to missing outcome data. LOW: outcome data available for nearly all participants OR appropriate methods handle missingness with sensitivity analyses showing the result is robust. SOME CONCERNS: missing data present but no positive evidence the result is biased. HIGH: missingness on the outcome plausibly depends on its true value AND missing fraction is comparable to event count (binary outcomes) or substantial (continuous outcomes); OR LOCF/single imputation used for substantial missingness. |
| rob2_measurement | LowSome concernsHigh | RoB 2 Domain 4: bias in measurement of the outcome. LOW: outcome is objective (all-cause mortality, automated lab values) AND ascertainment methods identical across groups, OR outcome assessors explicitly blinded to intervention. SOME CONCERNS: subjective outcome assessed by unblinded assessors in a neutral-belief context, OR blinding unclear for a judgment-dependent outcome. HIGH: subjective outcome (patient-reported, clinician-graded severity) assessed by unblinded assessors with strong prior beliefs about the intervention. |
| rob2_reporting | LowSome concernsHigh | RoB 2 Domain 5: bias in selection of the reported result. LOW: pre-registered protocol or SAP (dated before outcome unblinding) available AND reported numerical result matches the pre-specified primary analysis. SOME CONCERNS: no protocol available but no apparent selective outcome or analysis reporting. HIGH: evidence of selective outcome reporting, outcome switching, analysis switching, or selection of the most favorable of multiple analyses. |
| rob2_overall | LowSome concernsHigh | RoB 2 overall risk-of-bias judgment, derived from the five per-domain judgments per Cochrane Table 1. HIGH when any one domain is HIGH, OR when multiple domains are SOME CONCERNS and the combined concern substantially lowers confidence. SOME CONCERNS when at least one domain is SOME CONCERNS, no domain is HIGH, and the multi-domain escalation does not apply. LOW only when all five domains are LOW. |
| robins_confounding | LowModerateSeriousCritical | ROBINS-I Domain 1: bias from confounding in the intervention-vs-comparator effect estimate. LOW: design adds RCT-grade control (target-trial emulation with active comparator, negative-control falsification, or a quasi-experimental instrument) AND analysis covers important confounders. MODERATE: analysis adjusts for important measured confounders via propensity matching/weighting, regression, or stratification, but residual unmeasured confounding remains plausible. SERIOUS: the adjustment set omits a clinically-recognized driver of treatment choice or strong prognostic factor for the outcome, OR no quantitative adjustment was performed. CRITICAL: confounding renders the result uninterpretable. |
| robins_selection | LowModerateSeriousCritical | ROBINS-I Domain 2: bias from selection of participants into the analytic cohort. LOW: selection based on pre-intervention characteristics; analytic cohort defined at start of follow-up; no post-baseline exclusion correlated with intervention or outcome. MODERATE: some post-baseline exclusion or analytic-cohort filtering present but unlikely to depend on both intervention and outcome. SERIOUS: selection depends on both intervention and outcome (collider stratification, immortal-time bias, post-baseline cohort definition correlated with outcome). CRITICAL: selection virtually guarantees biased results. |
| robins_classification | LowModerateSeriousCritical | ROBINS-I Domain 3: bias from misclassification of intervention status. LOW: intervention status documented from the procedural record at the start of intervention. MODERATE: intervention status inferred from a proxy variable (administrative category, time-threshold cut-off, billing or ICD code) or from records recorded after intervention started; classification not influenced by outcome knowledge. SERIOUS: classification influenced by outcome knowledge (recall, retrospective recoding) or differential misclassification across groups. CRITICAL: classification errors invalidate the comparison. |
| robins_deviations | LowModerateSeriousCritical | ROBINS-I Domain 4: bias from deviations from intended interventions. LOW: no substantial deviations from intended intervention, OR the analysis appropriately estimates the targeted effect (ITT-style for assignment effect; g-methods for per-protocol effect). MODERATE: minor deviations not handled analytically. SERIOUS: substantial unbalanced deviations affecting the outcome, OR per-protocol analysis uses standard regression on time-varying confounders affected by prior intervention. CRITICAL: deviations render the comparison meaningless. |
| robins_missing | LowModerateSeriousCritical | ROBINS-I Domain 5: bias from missing data on intervention, outcome, or important confounders. LOW: complete data for at least 95% of participants with proportions similar across groups, OR appropriate multiple imputation, OR sensitivity analyses show results robust to missingness. MODERATE: some missing data with reasonable handling but incomplete documentation; missingness unlikely to depend on the true outcome value. SERIOUS: substantial missingness depending on the true outcome value or differing between groups, OR only LOCF / single imputation used for substantial missingness, OR no information on extent of missing data. CRITICAL: missing data so extensive and selective that the result cannot be interpreted. |
| robins_measurement | LowModerateSeriousCritical | ROBINS-I Domain 6: bias from outcome measurement (differential or non-differential measurement error). LOW: outcome is hard and objective (mortality, lab values, registry-linked death) AND ascertainment methods are identical across groups AND surveillance frequency does not differ by intervention. MODERATE: assessment methods comparable across groups; blinding not documented for a judgment-dependent outcome; minor ascertainment timing differences without documented detection differential. SERIOUS: differential ascertainment across groups (active vs passive surveillance, different diagnostic thresholds, different follow-up intensity), OR subjective outcome assessed by unblinded assessors with strong prior beliefs about the intervention. CRITICAL: measurement so unsuitable that the intervention-outcome relationship is uninterpretable. |
| robins_reporting | LowModerateSeriousCritical | ROBINS-I Domain 7: bias from selective reporting of the result. LOW: pre-registered protocol or statistical analysis plan available and results match the pre-specified analysis. MODERATE: no protocol available but no apparent selective outcome or analysis reporting. SERIOUS: evidence of selective outcome reporting, analysis switching, or subgroup-cherry-picking. CRITICAL: reporting selection virtually guarantees a biased result. |
| robins_overall | LowModerateSeriousCritical | ROBINS-I overall risk-of-bias judgment, computed as the worst per-domain judgment. LOW only when every domain is LOW (the result is comparable to a well-conducted RCT). MODERATE when at least one domain is MODERATE and no domain is SERIOUS or CRITICAL. SERIOUS when at least one domain is SERIOUS and no domain is CRITICAL. CRITICAL when at least one domain is CRITICAL. |