Interpreting the Forest Plot from the Shift Work Disorder Meta-Analysis
The forest plot from this systematic review demonstrates that shift work disorder affects 45.5% of nurses globally (95% CI: 39.2-51.8%), with substantial heterogeneity across studies (I² = 98.5%), indicating that prevalence estimates vary widely depending on diagnostic criteria, geographic region, and shift patterns 1.
Key Elements of the Forest Plot
Overall Pooled Prevalence
The diamond at the bottom of the forest plot represents the combined prevalence estimate of 45.5%, which is notably higher than the 26.5% prevalence found in general shift-working populations 2. This indicates nurses face disproportionately elevated risk compared to other occupational groups.
Heterogeneity Interpretation
The I² of 98.5% signals extreme variability between individual studies 1. This massive heterogeneity means:
- Individual study estimates scatter widely across the plot
- The confidence intervals of individual studies often don't overlap
- Pooled estimates must be interpreted cautiously
- Subgroup analyses become critical for understanding true prevalence
Critical Subgroup Findings Visible in the Plot
By Diagnostic Criteria:
- Studies using scales/questionnaires show 54.2% prevalence
- ICSD-3 criteria yield 40.2% prevalence
- ICSD-2 criteria yield 32.9% prevalence 1
This 21-percentage-point spread explains much of the heterogeneity—the forest plot will show studies clustering into distinct groups based on assessment method. Standardized diagnostic criteria (ICSD-3) provide more conservative and likely more accurate estimates 2.
By Geographic Region:
- African studies cluster at 51.0% prevalence
- Asian studies at 46.5% prevalence 1
These regional differences likely reflect variations in shift scheduling practices, nurse-to-patient ratios, and workplace support systems.
By Shift Pattern:
- Three-shift rotation models show the highest prevalence at 61.7% 1
- This represents a critical occupational hazard that should be visible as outlier estimates in the upper range of the forest plot
Understanding the Width of Confidence Intervals
Studies with:
- Narrow confidence intervals = larger sample sizes, more precise estimates
- Wide confidence intervals = smaller samples, less reliable estimates
The meta-regression found that larger sample sizes correlated with lower prevalence estimates 1, 2, suggesting potential selection bias in smaller studies that may oversample symptomatic nurses.
Clinical Implications from the Forest Plot Pattern
The forest plot visually demonstrates that shift work disorder is not a uniform phenomenon but varies dramatically based on:
Modifiable organizational factors: The 61.7% prevalence in three-shift models versus lower rates in other patterns indicates scheduling interventions can reduce risk 1
Assessment methodology: The 21-point prevalence difference between questionnaires and ICSD criteria means screening tools may overestimate true disorder prevalence, though they capture clinically significant symptoms 1, 2
Individual risk factors: Studies accounting for quick shifts (OR = 2.593), number of annual night shifts (OR = 1.010 per shift), anxiety (OR = 1.073), depression (OR = 1.178), and fatigue (OR = 1.145) will show higher prevalence estimates 1
Common Pitfalls in Interpretation
Do not assume the pooled estimate applies uniformly: A nurse working fixed day shifts in a well-staffed unit has vastly different risk than one rotating through three shifts with frequent quick returns. The forest plot's heterogeneity demands risk stratification.
Recognize publication bias: While funnel plots and Egger tests were performed 1, the predominance of studies from Africa and Asia may not represent global patterns, particularly in healthcare systems with different staffing models.
Distinguish screening from diagnosis: Higher prevalence with questionnaires versus ICSD criteria doesn't mean questionnaires are "wrong"—they capture the broader spectrum of circadian disruption and sleep disturbance that warrants intervention, even if not meeting full diagnostic thresholds 3, 1.