Version v01 | 2026-07-05
Prepared by Pier-Luc Langlais
OSF documentation version
This log documents inter-rater reliability monitoring during formal title and abstract screening. Screening is conducted in 20 successive non-overlapping blocks of approximately 5% of the corpus. Each block is assessed using independent pre-resolution decisions only.
Binary Fleiss’ kappa measures agreement on the operational decision to retain or exclude a reference. Nominal Fleiss’ kappa measures agreement across the full INC, MAYBE, and EXC decision framework.
The final PDF is accompanied by this RMarkdown source file and the CSV exports used for each block, providing a complete computational audit trail.
| Measure | Decision structure | Threshold | Protocol action |
|---|---|---|---|
| Binary Fleiss’ kappa | INC/MAYBE retained vs EXC | ≥ 0.75 | A single block below threshold triggers full recalibration. |
| Nominal Fleiss’ kappa | INC, MAYBE, and EXC | ≥ 0.60 | A single low block is recorded and monitored; three consecutive low blocks trigger targeted MAYBE recalibration. |
Binary kappa is treated as the primary safeguard because it measures agreement on whether a reference advances to full-text screening. Nominal kappa captures whether reviewers apply the full three-category framework consistently, particularly the MAYBE category.
Block 01 contained 341 imported references. Of these, 338 had exactly two independent pre-resolution decisions and were included in the inter-rater reliability analyses. The remaining 3 references had no recorded votes. Verification in Covidence confirmed that these records were residual duplicates removed during screening.
| Metric | N |
|---|---|
| References imported | 341 |
| References with exactly two independent votes | 338 |
| Residual duplicates verified in Covidence | 3 |
| References included in IRR calculations | 338 |
| Covidence # | Title |
|---|---|
| #139 | Transforming growth factor-β signaling: Tumorigenesis and targeting for cancer therapy |
| #439 | A Reasonable Officer: Examining the Relationships Among Stress, Training, and Performance in a Highly Realistic Lethal Force Scenario |
| #39 | Corrigendum to ‘Beyond Change Scores: Employing an Improved Statistical Approach to Analyze the Impact of Entry Fitness on Physical Performance During British Army Basic Training in Men and Women’ |
The nominal analysis treats INC, MAYBE, and EXC as three distinct categories. It therefore assesses agreement in the use of MAYBE as well as agreement on inclusion and exclusion decisions.
| Metric | Result |
|---|---|
| References analysed | 338 |
| Nominal Fleiss’ kappa | 0.591 |
| Raw agreement | 82.5% |
| Nominal disagreements | 59/338 |
| INC/MAYBE disagreements | 18/338 |
Interpretation: Below the pre-specified threshold; the use of MAYBE decisions is monitored.
# Three-category agreement
nominal_pairs <- compact_to_two_raters(block_decisions)
kappa_nominal <- irr::kappam.fleiss(
nominal_pairs,
exact = FALSE
)$value
The binary analysis collapses INC and MAYBE into a retained category, contrasted with EXC. It assesses agreement on whether each reference advances to full-text screening.
| Metric | Result |
|---|---|
| References analysed | 338 |
| Binary Fleiss’ kappa | 0.691 |
| Raw agreement | 87.9% |
| Binary conflicts | 41/338 |
| Conflict rate (95% Wilson CI) | 12.1% (9.1–16.0%) |
Interpretation: Below the pre-specified threshold; full recalibration is required.
# Binary recode: EXC = 0; INC and MAYBE = 1
binary_pairs <- nominal_pairs
binary_pairs[binary_pairs %in% c(1, 2)] <- 1
binary_pairs[binary_pairs == 0] <- 0
kappa_binary <- irr::kappam.fleiss(
binary_pairs,
exact = FALSE
)$value
The following measures are reported as complementary diagnostics of binary agreement during the initial monitoring phase. They do not replace the pre-specified binary Fleiss’ kappa threshold and do not independently determine the protocol action.
Binary Gwet’s AC1 provides an additional chance-corrected estimate of agreement. Retention-specific agreement describes agreement specifically on the operationally sensitive decision to retain a reference for full-text screening.
| Metric | Result |
|---|---|
| Binary Gwet’s AC1 | 0.800 |
| Retention-specific agreement | 77.3% |
Interpretation: These complementary measures are reported descriptively and diagnostically. No progression threshold is currently assigned to binary Gwet’s AC1 or retention-specific agreement. The pre-specified protocol action remains determined by binary and nominal Fleiss’ kappa.
The table below reports reference-level pre-resolution decision patterns among records with exactly two independent votes. A reference was classified as retained where both reviewers selected either INC or MAYBE. These results do not represent final post-arbitration outcomes.
| Pattern | N | Percentage |
|---|---|---|
| Excluded by both reviewers (EXC / EXC) | 227 | 67.2% |
| Conflicting retention decision (EXC / INC or MAYBE) | 41 | 12.1% |
| Retained by both reviewers (INC and/or MAYBE) | 70 | 20.7% |
Among references retained by both reviewers, the following nominal decision patterns were observed:
| Pattern | N | Percentage_of_retained |
|---|---|---|
| INC / INC | 26 | 37.1% |
| INC / MAYBE | 18 | 25.7% |
| MAYBE / MAYBE | 26 | 37.1% |
| Metric | Result |
|---|---|
| References imported | 341 |
| References with exactly two independent votes | 338 |
| References with zero votes | 3 |
| References excluded from IRR calculation | 3 |
| Binary Fleiss’ kappa | 0.691 |
| Nominal Fleiss’ kappa | 0.591 |
| Preceding low nominal-kappa blocks | 0 |
| Protocol action | Full recalibration |
Decision: Full recalibration required before the next block.
Rationale: Binary agreement fell below the pre-specified threshold of 0.75, indicating insufficient agreement on the retain-versus-exclude decision.
decision <- protocol_decision(
binary_kappa = kappa_binary,
nominal_kappa = kappa_nominal,
prior_low_nominal_blocks = PRIOR_CONSECUTIVE_LOW_NOMINAL_BLOCKS
)
Block 02 contained 341 imported references. Of these, 339 had exactly two independent pre-resolution decisions and were included in the inter-rater reliability analyses. The remaining 2 references had no recorded votes. Verification in Covidence confirmed that these records were residual duplicates removed during screening.
| Metric | N |
|---|---|
| References imported | 341 |
| References with exactly two independent votes | 339 |
| Residual duplicates verified in Covidence | 2 |
| References included in IRR calculations | 339 |
| Covidence # | Title |
|---|---|
| #643 | Bell 2015 : Effects of 13-Hour 20-Minute Work Shifts on Law Enforcement Officers’ Sleep, Cognitive Abilities, Health, Quality of Life, and Work Performance: The Phoenix Study |
| #768 | Bishop 1991: Limitations to heavy work at 21 degrees C of personnel wearing the U.S. military chemical defense ensemble |
The nominal analysis treats INC, MAYBE, and EXC as three distinct categories. It therefore assesses agreement in the use of MAYBE as well as agreement on inclusion and exclusion decisions.
| Metric | Result |
|---|---|
| References analysed | 339 |
| Nominal Fleiss’ kappa | 0.587 |
| Raw agreement | 81.1% |
| Nominal disagreements | 64/339 |
| INC/MAYBE disagreements | 15/339 |
Interpretation: Below the pre-specified threshold; the use of MAYBE decisions is monitored.
# Three-category agreement
nominal_pairs <- compact_to_two_raters(block_decisions)
kappa_nominal <- irr::kappam.fleiss(
nominal_pairs,
exact = FALSE
)$value
The binary analysis collapses INC and MAYBE into a retained category, contrasted with EXC. It assesses agreement on whether each reference advances to full-text screening.
| Metric | Result |
|---|---|
| References analysed | 339 |
| Binary Fleiss’ kappa | 0.651 |
| Raw agreement | 85.5% |
| Binary conflicts | 49/339 |
| Conflict rate (95% Wilson CI) | 14.5% (11.1–18.6%) |
Interpretation: Below the pre-specified threshold; full recalibration is required.
# Binary recode: EXC = 0; INC and MAYBE = 1
binary_pairs <- nominal_pairs
binary_pairs[binary_pairs %in% c(1, 2)] <- 1
binary_pairs[binary_pairs == 0] <- 0
kappa_binary <- irr::kappam.fleiss(
binary_pairs,
exact = FALSE
)$value
The following measures are reported as complementary diagnostics of binary agreement during the initial monitoring phase. They do not replace the pre-specified binary Fleiss’ kappa threshold and do not independently determine the protocol action.
Binary Gwet’s AC1 provides an additional chance-corrected estimate of agreement. Retention-specific agreement describes agreement specifically on the operationally sensitive decision to retain a reference for full-text screening.
| Metric | Result |
|---|---|
| Binary Gwet’s AC1 | 0.753 |
| Retention-specific agreement | 75.4% |
Interpretation: These complementary measures are reported descriptively and diagnostically. No progression threshold is currently assigned to binary Gwet’s AC1 or retention-specific agreement. The pre-specified protocol action remains determined by binary and nominal Fleiss’ kappa.
The table below reports reference-level pre-resolution decision patterns among records with exactly two independent votes. A reference was classified as retained where both reviewers selected either INC or MAYBE. These results do not represent final post-arbitration outcomes.
| Pattern | N | Percentage |
|---|---|---|
| Excluded by both reviewers (EXC / EXC) | 215 | 63.4% |
| Conflicting retention decision (EXC / INC or MAYBE) | 49 | 14.5% |
| Retained by both reviewers (INC and/or MAYBE) | 75 | 22.1% |
Among references retained by both reviewers, the following nominal decision patterns were observed:
| Pattern | N | Percentage_of_retained |
|---|---|---|
| INC / INC | 31 | 41.3% |
| INC / MAYBE | 15 | 20.0% |
| MAYBE / MAYBE | 29 | 38.7% |
| Metric | Result |
|---|---|
| References imported | 341 |
| References with exactly two independent votes | 339 |
| References with zero votes | 2 |
| References excluded from IRR calculation | 2 |
| Binary Fleiss’ kappa | 0.651 |
| Nominal Fleiss’ kappa | 0.587 |
| Preceding low nominal-kappa blocks | 0 |
| Protocol action | Full recalibration |
Decision: Full recalibration required before the next block.
Rationale: Binary agreement fell below the pre-specified threshold of 0.75, indicating insufficient agreement on the retain-versus-exclude decision.
decision <- protocol_decision(
binary_kappa = kappa_binary,
nominal_kappa = kappa_nominal,
prior_low_nominal_blocks = PRIOR_CONSECUTIVE_LOW_NOMINAL_BLOCKS
)
Block 03 contained 341 imported references. Of these, 338 had exactly two independent pre-resolution decisions and were included in the inter-rater reliability analyses. The remaining 3 references had no recorded votes. Verification in Covidence confirmed that these records were residual duplicates removed during screening.
| Metric | N |
|---|---|
| References imported | 341 |
| References with exactly two independent votes | 338 |
| Residual duplicates verified in Covidence | 3 |
| References included in IRR calculations | 338 |
| Covidence # | Title |
|---|---|
| #1155 | [Evaluation of Borg’s perceived exertion scale in cardiac rehabilitation] |
| #1252 | Differences between U.S. Army Trainees and Active Duty Soldiers in Performing Physically Demanding Occupational Tasks: 2950 Board #233 June 1 3:30 PM - 5:00 PM |
| #1295 | Systematic Review and Meta-Analysis: Iron Deficiency and Iron Deficiency Anemia Among Military Recruits Undergoing Basic Combat Training |
The nominal analysis treats INC, MAYBE, and EXC as three distinct categories. It therefore assesses agreement in the use of MAYBE as well as agreement on inclusion and exclusion decisions.
| Metric | Result |
|---|---|
| References analysed | 338 |
| Nominal Fleiss’ kappa | 0.669 |
| Raw agreement | 86.1% |
| Nominal disagreements | 47/338 |
| INC/MAYBE disagreements | 12/338 |
Interpretation: Substantial agreement; threshold met.
# Three-category agreement
nominal_pairs <- compact_to_two_raters(block_decisions)
kappa_nominal <- irr::kappam.fleiss(
nominal_pairs,
exact = FALSE
)$value
The binary analysis collapses INC and MAYBE into a retained category, contrasted with EXC. It assesses agreement on whether each reference advances to full-text screening.
| Metric | Result |
|---|---|
| References analysed | 338 |
| Binary Fleiss’ kappa | 0.734 |
| Raw agreement | 89.6% |
| Binary conflicts | 35/338 |
| Conflict rate (95% Wilson CI) | 10.4% (7.5–14.1%) |
Interpretation: Below the pre-specified threshold; full recalibration is required.
# Binary recode: EXC = 0; INC and MAYBE = 1
binary_pairs <- nominal_pairs
binary_pairs[binary_pairs %in% c(1, 2)] <- 1
binary_pairs[binary_pairs == 0] <- 0
kappa_binary <- irr::kappam.fleiss(
binary_pairs,
exact = FALSE
)$value
The following measures are reported as complementary diagnostics of binary agreement during the initial monitoring phase. They do not replace the pre-specified binary Fleiss’ kappa threshold and do not independently determine the protocol action.
Binary Gwet’s AC1 provides an additional chance-corrected estimate of agreement. Retention-specific agreement describes agreement specifically on the operationally sensitive decision to retain a reference for full-text screening.
| Metric | Result |
|---|---|
| Binary Gwet’s AC1 | 0.830 |
| Retention-specific agreement | 80.4% |
Interpretation: These complementary measures are reported descriptively and diagnostically. No progression threshold is currently assigned to binary Gwet’s AC1 or retention-specific agreement. The pre-specified protocol action remains determined by binary and nominal Fleiss’ kappa.
The table below reports reference-level pre-resolution decision patterns among records with exactly two independent votes. A reference was classified as retained where both reviewers selected either INC or MAYBE. These results do not represent final post-arbitration outcomes.
| Pattern | N | Percentage |
|---|---|---|
| Excluded by both reviewers (EXC / EXC) | 231 | 68.3% |
| Conflicting retention decision (EXC / INC or MAYBE) | 35 | 10.4% |
| Retained by both reviewers (INC and/or MAYBE) | 72 | 21.3% |
Among references retained by both reviewers, the following nominal decision patterns were observed:
| Pattern | N | Percentage_of_retained |
|---|---|---|
| INC / INC | 22 | 30.6% |
| INC / MAYBE | 12 | 16.7% |
| MAYBE / MAYBE | 38 | 52.8% |
| Metric | Result |
|---|---|
| References imported | 341 |
| References with exactly two independent votes | 338 |
| References with zero votes | 3 |
| References excluded from IRR calculation | 3 |
| Binary Fleiss’ kappa | 0.734 |
| Nominal Fleiss’ kappa | 0.669 |
| Preceding low nominal-kappa blocks | 0 |
| Protocol action | Full recalibration |
Decision: Full recalibration required before the next block.
Rationale: Binary agreement fell below the pre-specified threshold of 0.75, indicating insufficient agreement on the retain-versus-exclude decision.
decision <- protocol_decision(
binary_kappa = kappa_binary,
nominal_kappa = kappa_nominal,
prior_low_nominal_blocks = PRIOR_CONSECUTIVE_LOW_NOMINAL_BLOCKS
)
Block 04 contained 342 imported references. Of these, 342 had exactly two independent pre-resolution decisions and were included in the inter-rater reliability analyses. The remaining 0 references had no recorded votes. Verification in Covidence confirmed that these records were residual duplicates removed during screening.
| Metric | N |
|---|---|
| References imported | 342 |
| References with exactly two independent votes | 342 |
| Residual duplicates verified in Covidence | 0 |
| References included in IRR calculations | 342 |
The nominal analysis treats INC, MAYBE, and EXC as three distinct categories. It therefore assesses agreement in the use of MAYBE as well as agreement on inclusion and exclusion decisions.
| Metric | Result |
|---|---|
| References analysed | 342 |
| Nominal Fleiss’ kappa | 0.605 |
| Raw agreement | 82.5% |
| Nominal disagreements | 60/342 |
| INC/MAYBE disagreements | 23/342 |
Interpretation: Substantial agreement; threshold met.
# Three-category agreement
nominal_pairs <- compact_to_two_raters(block_decisions)
kappa_nominal <- irr::kappam.fleiss(
nominal_pairs,
exact = FALSE
)$value
The binary analysis collapses INC and MAYBE into a retained category, contrasted with EXC. It assesses agreement on whether each reference advances to full-text screening.
| Metric | Result |
|---|---|
| References analysed | 342 |
| Binary Fleiss’ kappa | 0.735 |
| Raw agreement | 89.2% |
| Binary conflicts | 37/342 |
| Conflict rate (95% Wilson CI) | 10.8% (8.0–14.6%) |
Interpretation: Below the pre-specified threshold; full recalibration is required.
# Binary recode: EXC = 0; INC and MAYBE = 1
binary_pairs <- nominal_pairs
binary_pairs[binary_pairs %in% c(1, 2)] <- 1
binary_pairs[binary_pairs == 0] <- 0
kappa_binary <- irr::kappam.fleiss(
binary_pairs,
exact = FALSE
)$value
The following measures are reported as complementary diagnostics of binary agreement during the initial monitoring phase. They do not replace the pre-specified binary Fleiss’ kappa threshold and do not independently determine the protocol action.
Binary Gwet’s AC1 provides an additional chance-corrected estimate of agreement. Retention-specific agreement describes agreement specifically on the operationally sensitive decision to retain a reference for full-text screening.
| Metric | Result |
|---|---|
| Binary Gwet’s AC1 | 0.817 |
| Retention-specific agreement | 81.0% |
Interpretation: These complementary measures are reported descriptively and diagnostically. No progression threshold is currently assigned to binary Gwet’s AC1 or retention-specific agreement. The pre-specified protocol action remains determined by binary and nominal Fleiss’ kappa.
The table below reports reference-level pre-resolution decision patterns among records with exactly two independent votes. A reference was classified as retained where both reviewers selected either INC or MAYBE. These results do not represent final post-arbitration outcomes.
| Pattern | N | Percentage |
|---|---|---|
| Excluded by both reviewers (EXC / EXC) | 226 | 66.1% |
| Conflicting retention decision (EXC / INC or MAYBE) | 37 | 10.8% |
| Retained by both reviewers (INC and/or MAYBE) | 79 | 23.1% |
Among references retained by both reviewers, the following nominal decision patterns were observed:
| Pattern | N | Percentage_of_retained |
|---|---|---|
| INC / INC | 19 | 24.1% |
| INC / MAYBE | 23 | 29.1% |
| MAYBE / MAYBE | 37 | 46.8% |
| Metric | Result |
|---|---|
| References imported | 342 |
| References with exactly two independent votes | 342 |
| References with zero votes | 0 |
| References excluded from IRR calculation | 0 |
| Binary Fleiss’ kappa | 0.735 |
| Nominal Fleiss’ kappa | 0.605 |
| Preceding low nominal-kappa blocks | 0 |
| Protocol action | Full recalibration |
Decision: Full recalibration required before the next block.
Rationale: Binary agreement fell below the pre-specified threshold of 0.75, indicating insufficient agreement on the retain-versus-exclude decision.
decision <- protocol_decision(
binary_kappa = kappa_binary,
nominal_kappa = kappa_nominal,
prior_low_nominal_blocks = PRIOR_CONSECUTIVE_LOW_NOMINAL_BLOCKS
)
Block 05 contained 342 imported references. Of these, 339 had exactly two independent pre-resolution decisions and were included in the inter-rater reliability analyses. The remaining 3 references had no recorded votes. Verification in Covidence confirmed that these records were residual duplicates removed during screening.
| Metric | N |
|---|---|
| References imported | 342 |
| References with exactly two independent votes | 339 |
| Residual duplicates verified in Covidence | 3 |
| References included in IRR calculations | 339 |
| Covidence # | Title |
|---|---|
| #2181 | Eastern district AAHPERD 2007 research keynote: Effects of current trends in youth fitness on physical performance assessments and training methodologies of the United States Military Academy and the Army |
| #2339 | Occupational group as a moderator of the job satisfaction-job performance relationship |
| #2412 | Making whole blood available in austere medical environments: donor performance and safety |
The nominal analysis treats INC, MAYBE, and EXC as three distinct categories. It therefore assesses agreement in the use of MAYBE as well as agreement on inclusion and exclusion decisions.
| Metric | Result |
|---|---|
| References analysed | 339 |
| Nominal Fleiss’ kappa | 0.516 |
| Raw agreement | 81.7% |
| Nominal disagreements | 62/339 |
| INC/MAYBE disagreements | 19/339 |
Interpretation: Below the pre-specified threshold; the use of MAYBE decisions is monitored.
# Three-category agreement
nominal_pairs <- compact_to_two_raters(block_decisions)
kappa_nominal <- irr::kappam.fleiss(
nominal_pairs,
exact = FALSE
)$value
The binary analysis collapses INC and MAYBE into a retained category, contrasted with EXC. It assesses agreement on whether each reference advances to full-text screening.
| Metric | Result |
|---|---|
| References analysed | 339 |
| Binary Fleiss’ kappa | 0.640 |
| Raw agreement | 87.3% |
| Binary conflicts | 43/339 |
| Conflict rate (95% Wilson CI) | 12.7% (9.6–16.6%) |
Interpretation: Below the pre-specified threshold; full recalibration is required.
# Binary recode: EXC = 0; INC and MAYBE = 1
binary_pairs <- nominal_pairs
binary_pairs[binary_pairs %in% c(1, 2)] <- 1
binary_pairs[binary_pairs == 0] <- 0
kappa_binary <- irr::kappam.fleiss(
binary_pairs,
exact = FALSE
)$value
The following measures are reported as complementary diagnostics of binary agreement during the initial monitoring phase. They do not replace the pre-specified binary Fleiss’ kappa threshold and do not independently determine the protocol action.
Binary Gwet’s AC1 provides an additional chance-corrected estimate of agreement. Retention-specific agreement describes agreement specifically on the operationally sensitive decision to retain a reference for full-text screening.
| Metric | Result |
|---|---|
| Binary Gwet’s AC1 | 0.804 |
| Retention-specific agreement | 72.3% |
Interpretation: These complementary measures are reported descriptively and diagnostically. No progression threshold is currently assigned to binary Gwet’s AC1 or retention-specific agreement. The pre-specified protocol action remains determined by binary and nominal Fleiss’ kappa.
The table below reports reference-level pre-resolution decision patterns among records with exactly two independent votes. A reference was classified as retained where both reviewers selected either INC or MAYBE. These results do not represent final post-arbitration outcomes.
| Pattern | N | Percentage |
|---|---|---|
| Excluded by both reviewers (EXC / EXC) | 240 | 70.8% |
| Conflicting retention decision (EXC / INC or MAYBE) | 43 | 12.7% |
| Retained by both reviewers (INC and/or MAYBE) | 56 | 16.5% |
Among references retained by both reviewers, the following nominal decision patterns were observed:
| Pattern | N | Percentage_of_retained |
|---|---|---|
| INC / INC | 16 | 28.6% |
| INC / MAYBE | 19 | 33.9% |
| MAYBE / MAYBE | 21 | 37.5% |
| Metric | Result |
|---|---|
| References imported | 342 |
| References with exactly two independent votes | 339 |
| References with zero votes | 3 |
| References excluded from IRR calculation | 3 |
| Binary Fleiss’ kappa | 0.640 |
| Nominal Fleiss’ kappa | 0.516 |
| Preceding low nominal-kappa blocks | 0 |
| Protocol action | Full recalibration |
Decision: Full recalibration required before the next block.
Rationale: Binary agreement fell below the pre-specified threshold of 0.75, indicating insufficient agreement on the retain-versus-exclude decision.
decision <- protocol_decision(
binary_kappa = kappa_binary,
nominal_kappa = kappa_nominal,
prior_low_nominal_blocks = PRIOR_CONSECUTIVE_LOW_NOMINAL_BLOCKS
)
Block 06 contained 339 imported references. Of these, 336 had exactly two independent pre-resolution decisions and were included in the inter-rater reliability analyses. The remaining 3 references had no recorded votes. Verification in Covidence confirmed that these records were residual duplicates removed during screening.
| Metric | N |
|---|---|
| References imported | 339 |
| References with exactly two independent votes | 336 |
| Residual duplicates verified in Covidence | 3 |
| References included in IRR calculations | 336 |
| Covidence # | Title |
|---|---|
| #2690 | Psychological, physical, and cognitive factors that influence tactical performance during a military relevant virtual reality scenario |
| #2693 | Psychological, Physical, And Cognitive Associations With Tactical Performance During A Military Relevant Virtual Reality Scenario: 1553…American College of Sports Medicine (ACSM) Annual Meeting, May 27-30, 2025, Atlanta, Georgia |
| #2709 | U.S. Army Physical Demands Study: Development of the Occupational Physical Assessment Test for Combat Arms soldiers |
The nominal analysis treats INC, MAYBE, and EXC as three distinct categories. It therefore assesses agreement in the use of MAYBE as well as agreement on inclusion and exclusion decisions.
| Metric | Result |
|---|---|
| References analysed | 336 |
| Nominal Fleiss’ kappa | 0.602 |
| Raw agreement | 84.5% |
| Nominal disagreements | 52/336 |
| INC/MAYBE disagreements | 16/336 |
Interpretation: Substantial agreement; threshold met.
# Three-category agreement
nominal_pairs <- compact_to_two_raters(block_decisions)
kappa_nominal <- irr::kappam.fleiss(
nominal_pairs,
exact = FALSE
)$value
The binary analysis collapses INC and MAYBE into a retained category, contrasted with EXC. It assesses agreement on whether each reference advances to full-text screening.
| Metric | Result |
|---|---|
| References analysed | 336 |
| Binary Fleiss’ kappa | 0.705 |
| Raw agreement | 89.3% |
| Binary conflicts | 36/336 |
| Conflict rate (95% Wilson CI) | 10.7% (7.8–14.5%) |
Interpretation: Below the pre-specified threshold; full recalibration is required.
# Binary recode: EXC = 0; INC and MAYBE = 1
binary_pairs <- nominal_pairs
binary_pairs[binary_pairs %in% c(1, 2)] <- 1
binary_pairs[binary_pairs == 0] <- 0
kappa_binary <- irr::kappam.fleiss(
binary_pairs,
exact = FALSE
)$value
The following measures are reported as complementary diagnostics of binary agreement during the initial monitoring phase. They do not replace the pre-specified binary Fleiss’ kappa threshold and do not independently determine the protocol action.
Binary Gwet’s AC1 provides an additional chance-corrected estimate of agreement. Retention-specific agreement describes agreement specifically on the operationally sensitive decision to retain a reference for full-text screening.
| Metric | Result |
|---|---|
| Binary Gwet’s AC1 | 0.832 |
| Retention-specific agreement | 77.5% |
Interpretation: These complementary measures are reported descriptively and diagnostically. No progression threshold is currently assigned to binary Gwet’s AC1 or retention-specific agreement. The pre-specified protocol action remains determined by binary and nominal Fleiss’ kappa.
The table below reports reference-level pre-resolution decision patterns among records with exactly two independent votes. A reference was classified as retained where both reviewers selected either INC or MAYBE. These results do not represent final post-arbitration outcomes.
| Pattern | N | Percentage |
|---|---|---|
| Excluded by both reviewers (EXC / EXC) | 238 | 70.8% |
| Conflicting retention decision (EXC / INC or MAYBE) | 36 | 10.7% |
| Retained by both reviewers (INC and/or MAYBE) | 62 | 18.5% |
Among references retained by both reviewers, the following nominal decision patterns were observed:
| Pattern | N | Percentage_of_retained |
|---|---|---|
| INC / INC | 17 | 27.4% |
| INC / MAYBE | 16 | 25.8% |
| MAYBE / MAYBE | 29 | 46.8% |
| Metric | Result |
|---|---|
| References imported | 339 |
| References with exactly two independent votes | 336 |
| References with zero votes | 3 |
| References excluded from IRR calculation | 3 |
| Binary Fleiss’ kappa | 0.705 |
| Nominal Fleiss’ kappa | 0.602 |
| Preceding low nominal-kappa blocks | 0 |
| Protocol action | Full recalibration |
Decision: Full recalibration required before the next block.
Rationale: Binary agreement fell below the pre-specified threshold of 0.75, indicating insufficient agreement on the retain-versus-exclude decision.
decision <- protocol_decision(
binary_kappa = kappa_binary,
nominal_kappa = kappa_nominal,
prior_low_nominal_blocks = PRIOR_CONSECUTIVE_LOW_NOMINAL_BLOCKS
)
Block 07 contained 341 imported references. Of these, 337 had exactly two independent pre-resolution decisions and were included in the inter-rater reliability analyses. The remaining 4 references had no recorded votes. Verification in Covidence confirmed that these records were residual duplicates removed during screening.
| Metric | N |
|---|---|
| References imported | 341 |
| References with exactly two independent votes | 337 |
| Residual duplicates verified in Covidence | 4 |
| References included in IRR calculations | 337 |
| Covidence # | Title |
|---|---|
| #2987 | Establishing A Cognitive, Health, Physical, And Social-emotional Toolkit To Predict Soldier Performance: 486 |
| #3211 | How Effective Is Initial Military-Specific Training in the Development of Physical Performance of Soldiers? |
| #3352 | United States Special Operations Command South Training Program’s Effectiveness: A Case Study |
| #3372 | Preliminary examination of timesharing in United States Air Force ROTC cadets |
The nominal analysis treats INC, MAYBE, and EXC as three distinct categories. It therefore assesses agreement in the use of MAYBE as well as agreement on inclusion and exclusion decisions.
| Metric | Result |
|---|---|
| References analysed | 337 |
| Nominal Fleiss’ kappa | 0.692 |
| Raw agreement | 87.5% |
| Nominal disagreements | 42/337 |
| INC/MAYBE disagreements | 14/337 |
Interpretation: Substantial agreement; threshold met.
# Three-category agreement
nominal_pairs <- compact_to_two_raters(block_decisions)
kappa_nominal <- irr::kappam.fleiss(
nominal_pairs,
exact = FALSE
)$value
The binary analysis collapses INC and MAYBE into a retained category, contrasted with EXC. It assesses agreement on whether each reference advances to full-text screening.
| Metric | Result |
|---|---|
| References analysed | 337 |
| Binary Fleiss’ kappa | 0.781 |
| Raw agreement | 91.7% |
| Binary conflicts | 28/337 |
| Conflict rate (95% Wilson CI) | 8.3% (5.8–11.7%) |
Interpretation: Excellent agreement; threshold met.
# Binary recode: EXC = 0; INC and MAYBE = 1
binary_pairs <- nominal_pairs
binary_pairs[binary_pairs %in% c(1, 2)] <- 1
binary_pairs[binary_pairs == 0] <- 0
kappa_binary <- irr::kappam.fleiss(
binary_pairs,
exact = FALSE
)$value
The following measures are reported as complementary diagnostics of binary agreement during the initial monitoring phase. They do not replace the pre-specified binary Fleiss’ kappa threshold and do not independently determine the protocol action.
Binary Gwet’s AC1 provides an additional chance-corrected estimate of agreement. Retention-specific agreement describes agreement specifically on the operationally sensitive decision to retain a reference for full-text screening.
| Metric | Result |
|---|---|
| Binary Gwet’s AC1 | 0.866 |
| Retention-specific agreement | 83.7% |
Interpretation: These complementary measures are reported descriptively and diagnostically. No progression threshold is currently assigned to binary Gwet’s AC1 or retention-specific agreement. The pre-specified protocol action remains determined by binary and nominal Fleiss’ kappa.
The table below reports reference-level pre-resolution decision patterns among records with exactly two independent votes. A reference was classified as retained where both reviewers selected either INC or MAYBE. These results do not represent final post-arbitration outcomes.
| Pattern | N | Percentage |
|---|---|---|
| Excluded by both reviewers (EXC / EXC) | 237 | 70.3% |
| Conflicting retention decision (EXC / INC or MAYBE) | 28 | 8.3% |
| Retained by both reviewers (INC and/or MAYBE) | 72 | 21.4% |
Among references retained by both reviewers, the following nominal decision patterns were observed:
| Pattern | N | Percentage_of_retained |
|---|---|---|
| INC / INC | 13 | 18.1% |
| INC / MAYBE | 14 | 19.4% |
| MAYBE / MAYBE | 45 | 62.5% |
| Metric | Result |
|---|---|
| References imported | 341 |
| References with exactly two independent votes | 337 |
| References with zero votes | 4 |
| References excluded from IRR calculation | 4 |
| Binary Fleiss’ kappa | 0.781 |
| Nominal Fleiss’ kappa | 0.692 |
| Preceding low nominal-kappa blocks | 0 |
| Protocol action | Continue |
Decision: Screening may continue without recalibration.
Rationale: Both pre-specified agreement thresholds were met.
decision <- protocol_decision(
binary_kappa = kappa_binary,
nominal_kappa = kappa_nominal,
prior_low_nominal_blocks = PRIOR_CONSECUTIVE_LOW_NOMINAL_BLOCKS
)
Block 08 contained 342 imported references. Of these, 339 had exactly two independent pre-resolution decisions and were included in the inter-rater reliability analyses. The remaining 3 references had no recorded votes. Verification in Covidence confirmed that these records were residual duplicates removed during screening.
| Metric | N |
|---|---|
| References imported | 342 |
| References with exactly two independent votes | 339 |
| Residual duplicates verified in Covidence | 3 |
| References included in IRR calculations | 339 |
| Covidence # | Title |
|---|---|
| 3440 | Heart Rate Reserve: An Objective Measure of Soldiers’ Physical Exertion During Field Operations: 2955 Board #238 June 1 3:30 PM - 5:00 PM |
| #3568 | Predictors Of Physical Readiness In Rotary Wing Naval Aviators: 1479 |
| #3647 | β-Alanine supplementation and military performance |
The nominal analysis treats INC, MAYBE, and EXC as three distinct categories. It therefore assesses agreement in the use of MAYBE as well as agreement on inclusion and exclusion decisions.
| Metric | Result |
|---|---|
| References analysed | 339 |
| Nominal Fleiss’ kappa | 0.628 |
| Raw agreement | 85.0% |
| Nominal disagreements | 51/339 |
| INC/MAYBE disagreements | 13/339 |
Interpretation: Substantial agreement; threshold met.
# Three-category agreement
nominal_pairs <- compact_to_two_raters(block_decisions)
kappa_nominal <- irr::kappam.fleiss(
nominal_pairs,
exact = FALSE
)$value
The binary analysis collapses INC and MAYBE into a retained category, contrasted with EXC. It assesses agreement on whether each reference advances to full-text screening.
| Metric | Result |
|---|---|
| References analysed | 339 |
| Binary Fleiss’ kappa | 0.702 |
| Raw agreement | 88.8% |
| Binary conflicts | 38/339 |
| Conflict rate (95% Wilson CI) | 11.2% (8.3–15.0%) |
Interpretation: Below the pre-specified threshold; full recalibration is required.
# Binary recode: EXC = 0; INC and MAYBE = 1
binary_pairs <- nominal_pairs
binary_pairs[binary_pairs %in% c(1, 2)] <- 1
binary_pairs[binary_pairs == 0] <- 0
kappa_binary <- irr::kappam.fleiss(
binary_pairs,
exact = FALSE
)$value
The following measures are reported as complementary diagnostics of binary agreement during the initial monitoring phase. They do not replace the pre-specified binary Fleiss’ kappa threshold and do not independently determine the protocol action.
Binary Gwet’s AC1 provides an additional chance-corrected estimate of agreement. Retention-specific agreement describes agreement specifically on the operationally sensitive decision to retain a reference for full-text screening.
| Metric | Result |
|---|---|
| Binary Gwet’s AC1 | 0.820 |
| Retention-specific agreement | 77.6% |
Interpretation: These complementary measures are reported descriptively and diagnostically. No progression threshold is currently assigned to binary Gwet’s AC1 or retention-specific agreement. The pre-specified protocol action remains determined by binary and nominal Fleiss’ kappa.
The table below reports reference-level pre-resolution decision patterns among records with exactly two independent votes. A reference was classified as retained where both reviewers selected either INC or MAYBE. These results do not represent final post-arbitration outcomes.
| Pattern | N | Percentage |
|---|---|---|
| Excluded by both reviewers (EXC / EXC) | 235 | 69.3% |
| Conflicting retention decision (EXC / INC or MAYBE) | 38 | 11.2% |
| Retained by both reviewers (INC and/or MAYBE) | 66 | 19.5% |
Among references retained by both reviewers, the following nominal decision patterns were observed:
| Pattern | N | Percentage_of_retained |
|---|---|---|
| INC / INC | 19 | 28.8% |
| INC / MAYBE | 13 | 19.7% |
| MAYBE / MAYBE | 34 | 51.5% |
| Metric | Result |
|---|---|
| References imported | 342 |
| References with exactly two independent votes | 339 |
| References with zero votes | 3 |
| References excluded from IRR calculation | 3 |
| Binary Fleiss’ kappa | 0.702 |
| Nominal Fleiss’ kappa | 0.628 |
| Preceding low nominal-kappa blocks | 0 |
| Protocol action | Full recalibration |
Decision: Full recalibration required before the next block.
Rationale: Binary agreement fell below the pre-specified threshold of 0.75, indicating insufficient agreement on the retain-versus-exclude decision.
decision <- protocol_decision(
binary_kappa = kappa_binary,
nominal_kappa = kappa_nominal,
prior_low_nominal_blocks = PRIOR_CONSECUTIVE_LOW_NOMINAL_BLOCKS
)
Block 09 contained 340 imported references. Of these, 338 had exactly two independent pre-resolution decisions and were included in the inter-rater reliability analyses. The remaining 2 references had no recorded votes. Verification in Covidence confirmed that these records were residual duplicates removed during screening.
| Metric | N |
|---|---|
| References imported | 340 |
| References with exactly two independent votes | 338 |
| Residual duplicates verified in Covidence | 2 |
| References included in IRR calculations | 338 |
| Covidence # | Title |
|---|---|
| 3896 | Key Factors Influencing Purchasing Of Dietary Supplements Among Female Service Academy Cadets |
| #3961 | The Benefit of Mental Skills Training on Performance and Stress Response in Military Personnel |
The nominal analysis treats INC, MAYBE, and EXC as three distinct categories. It therefore assesses agreement in the use of MAYBE as well as agreement on inclusion and exclusion decisions.
| Metric | Result |
|---|---|
| References analysed | 338 |
| Nominal Fleiss’ kappa | 0.630 |
| Raw agreement | 84.9% |
| Nominal disagreements | 51/338 |
| INC/MAYBE disagreements | 18/338 |
Interpretation: Substantial agreement; threshold met.
# Three-category agreement
nominal_pairs <- compact_to_two_raters(block_decisions)
kappa_nominal <- irr::kappam.fleiss(
nominal_pairs,
exact = FALSE
)$value
The binary analysis collapses INC and MAYBE into a retained category, contrasted with EXC. It assesses agreement on whether each reference advances to full-text screening.
| Metric | Result |
|---|---|
| References analysed | 338 |
| Binary Fleiss’ kappa | 0.742 |
| Raw agreement | 90.2% |
| Binary conflicts | 33/338 |
| Conflict rate (95% Wilson CI) | 9.8% (7.0–13.4%) |
Interpretation: Below the pre-specified threshold; full recalibration is required.
# Binary recode: EXC = 0; INC and MAYBE = 1
binary_pairs <- nominal_pairs
binary_pairs[binary_pairs %in% c(1, 2)] <- 1
binary_pairs[binary_pairs == 0] <- 0
kappa_binary <- irr::kappam.fleiss(
binary_pairs,
exact = FALSE
)$value
The following measures are reported as complementary diagnostics of binary agreement during the initial monitoring phase. They do not replace the pre-specified binary Fleiss’ kappa threshold and do not independently determine the protocol action.
Binary Gwet’s AC1 provides an additional chance-corrected estimate of agreement. Retention-specific agreement describes agreement specifically on the operationally sensitive decision to retain a reference for full-text screening.
| Metric | Result |
|---|---|
| Binary Gwet’s AC1 | 0.843 |
| Retention-specific agreement | 80.7% |
Interpretation: These complementary measures are reported descriptively and diagnostically. No progression threshold is currently assigned to binary Gwet’s AC1 or retention-specific agreement. The pre-specified protocol action remains determined by binary and nominal Fleiss’ kappa.
The table below reports reference-level pre-resolution decision patterns among records with exactly two independent votes. A reference was classified as retained where both reviewers selected either INC or MAYBE. These results do not represent final post-arbitration outcomes.
| Pattern | N | Percentage |
|---|---|---|
| Excluded by both reviewers (EXC / EXC) | 236 | 69.8% |
| Conflicting retention decision (EXC / INC or MAYBE) | 33 | 9.8% |
| Retained by both reviewers (INC and/or MAYBE) | 69 | 20.4% |
Among references retained by both reviewers, the following nominal decision patterns were observed:
| Pattern | N | Percentage_of_retained |
|---|---|---|
| INC / INC | 20 | 29.0% |
| INC / MAYBE | 18 | 26.1% |
| MAYBE / MAYBE | 31 | 44.9% |
| Metric | Result |
|---|---|
| References imported | 340 |
| References with exactly two independent votes | 338 |
| References with zero votes | 2 |
| References excluded from IRR calculation | 2 |
| Binary Fleiss’ kappa | 0.742 |
| Nominal Fleiss’ kappa | 0.630 |
| Preceding low nominal-kappa blocks | 0 |
| Protocol action | Full recalibration |
Decision: Full recalibration required before the next block.
Rationale: Binary agreement fell below the pre-specified threshold of 0.75, indicating insufficient agreement on the retain-versus-exclude decision.
decision <- protocol_decision(
binary_kappa = kappa_binary,
nominal_kappa = kappa_nominal,
prior_low_nominal_blocks = PRIOR_CONSECUTIVE_LOW_NOMINAL_BLOCKS
)
Block 10 contained 345 imported references. Of these, 343 had exactly two independent pre-resolution decisions and were included in the inter-rater reliability analyses. The remaining 2 references had no recorded votes. Verification in Covidence confirmed that these records were residual duplicates removed during screening.
| Metric | N |
|---|---|
| References imported | 345 |
| References with exactly two independent votes | 343 |
| Residual duplicates verified in Covidence | 2 |
| References included in IRR calculations | 343 |
| Covidence # | Title |
|---|---|
| 4795 | Effects of low speed wind on the recognition/identification and pass-through communication tasks of auditory situation awareness afforded by military hearing protection/enhancement devices and tactical communication and protective systems |
| #4804 | Novel Cooling Strategies for Military Training and Operations |
The nominal analysis treats INC, MAYBE, and EXC as three distinct categories. It therefore assesses agreement in the use of MAYBE as well as agreement on inclusion and exclusion decisions.
| Metric | Result |
|---|---|
| References analysed | 343 |
| Nominal Fleiss’ kappa | 0.620 |
| Raw agreement | 83.1% |
| Nominal disagreements | 58/343 |
| INC/MAYBE disagreements | 26/343 |
Interpretation: Substantial agreement; threshold met.
# Three-category agreement
nominal_pairs <- compact_to_two_raters(block_decisions)
kappa_nominal <- irr::kappam.fleiss(
nominal_pairs,
exact = FALSE
)$value
The binary analysis collapses INC and MAYBE into a retained category, contrasted with EXC. It assesses agreement on whether each reference advances to full-text screening.
| Metric | Result |
|---|---|
| References analysed | 343 |
| Binary Fleiss’ kappa | 0.773 |
| Raw agreement | 90.7% |
| Binary conflicts | 32/343 |
| Conflict rate (95% Wilson CI) | 9.3% (6.7–12.9%) |
Interpretation: Excellent agreement; threshold met.
# Binary recode: EXC = 0; INC and MAYBE = 1
binary_pairs <- nominal_pairs
binary_pairs[binary_pairs %in% c(1, 2)] <- 1
binary_pairs[binary_pairs == 0] <- 0
kappa_binary <- irr::kappam.fleiss(
binary_pairs,
exact = FALSE
)$value
The following measures are reported as complementary diagnostics of binary agreement during the initial monitoring phase. They do not replace the pre-specified binary Fleiss’ kappa threshold and do not independently determine the protocol action.
Binary Gwet’s AC1 provides an additional chance-corrected estimate of agreement. Retention-specific agreement describes agreement specifically on the operationally sensitive decision to retain a reference for full-text screening.
| Metric | Result |
|---|---|
| Binary Gwet’s AC1 | 0.842 |
| Retention-specific agreement | 83.8% |
Interpretation: These complementary measures are reported descriptively and diagnostically. No progression threshold is currently assigned to binary Gwet’s AC1 or retention-specific agreement. The pre-specified protocol action remains determined by binary and nominal Fleiss’ kappa.
The table below reports reference-level pre-resolution decision patterns among records with exactly two independent votes. A reference was classified as retained where both reviewers selected either INC or MAYBE. These results do not represent final post-arbitration outcomes.
| Pattern | N | Percentage |
|---|---|---|
| Excluded by both reviewers (EXC / EXC) | 228 | 66.5% |
| Conflicting retention decision (EXC / INC or MAYBE) | 32 | 9.3% |
| Retained by both reviewers (INC and/or MAYBE) | 83 | 24.2% |
Among references retained by both reviewers, the following nominal decision patterns were observed:
| Pattern | N | Percentage_of_retained |
|---|---|---|
| INC / INC | 16 | 19.3% |
| INC / MAYBE | 26 | 31.3% |
| MAYBE / MAYBE | 41 | 49.4% |
| Metric | Result |
|---|---|
| References imported | 345 |
| References with exactly two independent votes | 343 |
| References with zero votes | 2 |
| References excluded from IRR calculation | 2 |
| Binary Fleiss’ kappa | 0.773 |
| Nominal Fleiss’ kappa | 0.620 |
| Preceding low nominal-kappa blocks | 0 |
| Protocol action | Continue |
Decision: Screening may continue without recalibration.
Rationale: Both pre-specified agreement thresholds were met.
decision <- protocol_decision(
binary_kappa = kappa_binary,
nominal_kappa = kappa_nominal,
prior_low_nominal_blocks = PRIOR_CONSECUTIVE_LOW_NOMINAL_BLOCKS
)
This midpoint analysis provides a retrospective descriptive and exploratory assessment of inter-rater reliability and screening uncertainty across the first 50% of the corpus. It does not modify the pre-specified progression rules or retrospectively alter any block-level decision. Binary Fleiss’ kappa remains the primary operational safeguard. Gwet’s AC1 and retention-specific agreement are examined as complementary diagnostics, consistent with the monitoring strategy established during the initial screening phase.
Where additional measures are introduced at the midpoint, including exclusion-specific agreement and the retention–exclusion specific-agreement gap, they are treated as retrospective descriptive diagnostics only.
The first-half monitoring profile is presented below for Blocks 01–10. Block-level estimates are retained to show fluctuations across the screening sequence. Binary Fleiss’ kappa remains the primary operational monitoring measure, while nominal kappa, Gwet’s AC1, retention-specific agreement, and conflict rates are reported as complementary diagnostics.
| Block | N | Nominal κ | Binary κ | Gwet AC1 | Retention-specific | Binary agreement | Binary conflicts |
|---|---|---|---|---|---|---|---|
| 01 | 338 | 0.591 | 0.691 | 0.800 | 77.3% | 87.9% | 41 (12.1%) |
| 02 | 339 | 0.587 | 0.651 | 0.753 | 75.4% | 85.5% | 49 (14.5%) |
| 03 | 338 | 0.669 | 0.734 | 0.830 | 80.4% | 89.6% | 35 (10.4%) |
| 04 | 342 | 0.605 | 0.735 | 0.817 | 81.0% | 89.2% | 37 (10.8%) |
| 05 | 339 | 0.516 | 0.640 | 0.804 | 72.3% | 87.3% | 43 (12.7%) |
| 06 | 336 | 0.602 | 0.705 | 0.832 | 77.5% | 89.3% | 36 (10.7%) |
| 07 | 337 | 0.692 | 0.781 | 0.866 | 83.7% | 91.7% | 28 (8.3%) |
| 08 | 339 | 0.628 | 0.702 | 0.820 | 77.6% | 88.8% | 38 (11.2%) |
| 09 | 338 | 0.630 | 0.742 | 0.843 | 80.7% | 90.2% | 33 (9.8%) |
| 10 | 343 | 0.620 | 0.773 | 0.842 | 83.8% | 90.7% | 32 (9.3%) |
Because overall binary agreement may be driven disproportionately by references jointly excluded by both reviewers, decision-specific agreement was examined separately for retained and excluded references. The retention-specific measure reflects agreement on the decision to retain a reference, whereas exclusion-specific agreement provides the corresponding measure for exclusion. Their difference is reported descriptively and is not used as an additional progression threshold.
| Block | Retention-specific | Exclusion-specific | Excl. - Ret. gap | Both retained | Both excluded | INC/MAYBE disagreement | MAYBE annotations / 100 refs |
|---|---|---|---|---|---|---|---|
| 01 | 77.3% | 91.7% | +14.4 pp | 70 (20.7%) | 227 (67.2%) | 18 (5.3%) | 29.6 |
| 02 | 75.4% | 89.8% | +14.4 pp | 75 (22.1%) | 215 (63.4%) | 15 (4.4%) | 31.0 |
| 03 | 80.4% | 93.0% | +12.5 pp | 72 (21.3%) | 231 (68.3%) | 12 (3.6%) | 32.2 |
| 04 | 81.0% | 92.4% | +11.4 pp | 79 (23.1%) | 226 (66.1%) | 23 (6.7%) | 36.3 |
| 05 | 72.3% | 91.8% | +19.5 pp | 56 (16.5%) | 240 (70.8%) | 19 (5.6%) | 25.4 |
| 06 | 77.5% | 93.0% | +15.5 pp | 62 (18.5%) | 238 (70.8%) | 16 (4.8%) | 29.8 |
| 07 | 83.7% | 94.4% | +10.7 pp | 72 (21.4%) | 237 (70.3%) | 14 (4.2%) | 37.1 |
| 08 | 77.6% | 92.5% | +14.9 pp | 66 (19.5%) | 235 (69.3%) | 13 (3.8%) | 33.0 |
| 09 | 80.7% | 93.5% | +12.8 pp | 69 (20.4%) | 236 (69.8%) | 18 (5.3%) | 31.7 |
| 10 | 83.8% | 93.4% | +9.6 pp | 83 (24.2%) | 228 (66.5%) | 26 (7.6%) | 40.8 |
To characterise the evolution of agreement during the first half of screening, Blocks 01–05 and Blocks 06–10 were compared as two successive 25% screening periods. Agreement coefficients were recalculated from the pooled pre-resolution reviewer decisions rather than obtained by averaging block-level estimates. Reference-level bootstrap 95% confidence intervals are reported for the principal agreement coefficients. Differences between periods are interpreted descriptively and do not constitute a formal estimate of the effect of recalibration or guide revision.
| Metric | Blocks 01–05 | Blocks 06–10 | Change | Blocks 01–10 |
|---|---|---|---|---|
| References analysed | 1,696 | 1,693 | — | 3,389 |
| Nominal Fleiss’ kappa | 0.596 [0.554–0.636] | 0.635 [0.597–0.671] | +0.039 | 0.615 [0.589–0.643] |
| Binary Fleiss’ kappa | 0.692 [0.648–0.729] | 0.742 [0.706–0.779] | +0.050 | 0.717 [0.689–0.743] |
| Gwet’s binary AC1 | 0.801 [0.772–0.829] | 0.840 [0.816–0.865] | +0.039 | 0.821 [0.803–0.839] |
| AC1 - binary kappa gap | 0.109 | 0.098 | -0.011 | 0.104 |
| Binary raw agreement | 87.9% | 90.1% | +2.2 pp | 89.0% |
| Binary conflict rate | 12.1% | 9.9% | -2.2 pp | 11.0% |
| Retention-specific agreement | 77.4% [74.2–80.3] | 80.8% [78.0–83.6] | +3.4 pp | 79.1% [77.0–81.1] |
| Exclusion-specific agreement | 91.7% [90.6–92.9] | 93.4% [92.4–94.4] | +1.6 pp | 92.6% [91.8–93.3] |
| Exclusion - retention gap | 14.3 pp | 12.5 pp | -1.8 pp | 13.5 pp |
| INC/MAYBE disagreement rate | 5.1% | 5.1% | +0.0 pp | 5.1% |
| MAYBE annotations / 100 references | 30.9 | 34.5 | +3.6 | 32.7 |
Two substantive revisions of the screening guide occurred during the first half of screening. Guide V01 was used for Blocks 01–02. Following recalibration after Block 02, Guide V02 was introduced and used for Blocks 03–05. A second substantive revision following Block 05 resulted in Guide V03, which was used for Blocks 06–10. Other recalibration episodes during this period did not result in a new guide version. Guide-version periods are shown in the longitudinal figures for descriptive context and are not interpreted as experimental intervention periods.
Block-level agreement coefficients were examined longitudinally to assess the evolution of reviewer agreement across the first 50% of screening. The prespecified operational thresholds are displayed for binary and nominal Fleiss’ kappa. Gwet’s AC1 is shown as a complementary chance-corrected diagnostic and has no independent progression threshold.
Note. Binary Fleiss’ κ remains the primary operational
monitoring statistic. Gwet’s AC1 is complementary and has no independent
progression threshold. Vertical lines indicate transitions from Guide
V01 to V02 after Block 02 and from Guide V02 to V03 after Block 05.
Retention-specific and exclusion-specific agreement were examined separately to determine whether high overall binary agreement was primarily driven by shared exclusion decisions. In accordance with the monitoring plan, retention-specific agreement is also examined at the midpoint as a potential candidate for prospective operational interpretation. No threshold was prespecified for this measure; therefore, its ability to distinguish blocks associated with substantive guide revision from those not requiring such revision is evaluated separately below.
Note. Retention-specific agreement reflects concordance on
retaining references, whereas exclusion-specific agreement reflects
concordance on exclusion. Vertical lines indicate transitions from Guide
V01 to V02 after Block 02 and from Guide V02 to V03 after Block 05. No
retention-specific progression threshold was prespecified.
The observation phase specified for retention-specific agreement was completed at Block 10. Blocks that triggered a full recalibration were therefore examined according to the documented outcome of the recalibration meeting. The primary classification follows the broader operational definition of substantive recalibration, including the introduction of numbered operational clarifications or a new Guide version. A sensitivity analysis applies the narrower criterion of whether the meeting resulted in a new Guide version.
Blocks 07 and 10 are not included in this comparison because they met the prespecified progression criteria and did not trigger a full recalibration meeting. Candidate cut-points are exploratory and are not applied retrospectively.
| Block | Guide | Binary κ | Retention-specific | Expanded outcome | Guide-version sensitivity |
|---|---|---|---|---|---|
| 01 | Guide V01 | 0.691 | 77.3% | Substantive recalibration | No guide revision |
| 02 | Guide V01 | 0.651 | 75.4% | Substantive recalibration | Guide revision |
| 03 | Guide V02 | 0.734 | 80.4% | Reiteration-only recalibration | No guide revision |
| 04 | Guide V02 | 0.735 | 81.0% | Reiteration-only recalibration | No guide revision |
| 05 | Guide V02 | 0.640 | 72.3% | Substantive recalibration | Guide revision |
| 06 | Guide V03 | 0.705 | 77.5% | Reiteration-only recalibration | No guide revision |
| 08 | Guide V03 | 0.702 | 77.6% | Reiteration-only recalibration | No guide revision |
| 09 | Guide V03 | 0.742 | 80.7% | Reiteration-only recalibration | No guide revision |
| Scenario | Substantive n | Reiteration n | Substantive range | Reiteration range | Separation gap | Complete separation | Candidate cut-point | Exact Wilcoxon p |
|---|---|---|---|---|---|---|---|---|
| Expanded operational-change definition | 3 | 5 | 72.3–77.3% | 77.5–81.0% | +0.2 pp | Yes | 77.4% | 0.036 |
| Guide-version-only sensitivity | 2 | 6 | 72.3–75.4% | 77.3–81.0% | +2.0 pp | Yes | 76.4% | 0.071 |
Note. The dashed line is a data-derived exploratory cut-point rather than a prespecified progression threshold. The exact Wilcoxon test is reported only as a rank-based descriptive diagnostic because the number of recalibration episodes, particularly substantive episodes, is very small. The sensitivity analysis examines whether the apparent separation depends on defining substantive change by Guide-version revision alone.
MAYBE-note annotations were examined as an operational record of uncertainty encountered during screening. The unit of analysis is the reviewer annotation, not the individual reference. Accordingly, annotation counts should not be interpreted as counts of unique MAYBE references.
Because the Sel-crit category was introduced prospectively beginning with Block 06 and Blocks 01–05 were not retrospectively recoded, the complete category composition cannot be compared directly between the two periods. The principal period comparison therefore uses only categories available throughout Blocks 01–10. Sel-crit is retained separately as a prospective monitoring category.
The Sel-crit category accounted for 89 annotations during Blocks 06–10 (15.2% of the observed MAYBE-note annotations in that period). No corresponding percentage is calculated for Blocks 01–05 because the category was not yet available.
| Category | Description | Blocks 01–05 | Blocks 06–10 | Change |
|---|---|---|---|---|
| Oper | operational performance indicator not operationalised | 166 (31.7%) | 134 (27.1%) | -4.6 pp |
| Gen-task | generic test/task versus tactical anchoring | 110 (21.0%) | 104 (21.0%) | +0.0 pp |
| Ctx-org | contextual/organisational performance | 96 (18.3%) | 91 (18.4%) | +0.1 pp |
| Pop | ambiguous population or civilian proxy | 65 (12.4%) | 98 (19.8%) | +7.4 pp |
| Abs | abstract unavailable or incomplete | 32 (6.1%) | 23 (4.6%) | -1.5 pp |
| Doc | uncertain document type | 21 (4.0%) | 27 (5.5%) | +1.4 pp |
| Admin | administrative productivity as a proxy | 15 (2.9%) | 5 (1.0%) | -1.9 pp |
| Equip | personal/corrective equipment | 12 (2.3%) | 7 (1.4%) | -0.9 pp |
| Rehab | recovery/prevention versus performance | 7 (1.3%) | 6 (1.2%) | -0.1 pp |
The harmonised comparison removes Sel-crit from both periods so that the denominator is based on the same category framework. Differences are descriptive and may reflect both evolving application of the screening guidance and variation in the references encountered.
Note. Values represent the percentage of categorised MAYBE-note annotations within each block. NC = not coded. Sel-crit was introduced prospectively at Block 06 and was not retrospectively assigned to Blocks 01–05. Vertical lines indicate transitions from Guide V01 to V02 and from Guide V02 to V03.
A small set of block-level associations was examined to assess whether the complementary diagnostics behaved in an interpretable manner across Blocks 01–10. These analyses are exploratory. With only ten blocks, they are not intended as confirmatory hypothesis tests, and p-values are reported only as descriptive aids.
Several of the agreement measures are derived from the same underlying decision table and are therefore mathematically related. Associations among them must not be interpreted as independent evidence. MAYBE-note associations may additionally reflect differences in block composition.
| Association | N | Spearman ρ | Descriptive p |
|---|---|---|---|
| Retained-vote share vs AC1 - binary κ gap | 10 | -0.697 | 0.025 |
| MAYBE annotation burden vs nominal κ | 10 | +0.697 | 0.025 |
| MAYBE annotation burden vs retention-specific agreement | 10 | +0.915 | <0.001 |
| Gen-task annotations / 100 references vs binary κ | 10 | -0.321 | 0.365 |
Note. The figure is diagnostic rather than inferential. A larger AC1–κ gap at lower retained-decision prevalence would be consistent with the intended use of AC1 as a complementary diagnostic of the influence of decision distribution on κ. It would not establish AC1 as a superior or more accurate agreement coefficient.