Shortening of the Pittsburgh Sleep Quality Index Survey Using Factor Analysis

Abstract

Objective/introduction: Lengthy surveys have the potential to burden users and can lead to inaccuracies. Conducting analyses to shorten existing validated surveys is beneficial. The objective, therefore, was to shorten the Pittsburgh Quality Sleep Index (PSQI) for young adults.

Methods: PSQI data from 1246 college students were used. An exploratory factor analysis (FA) was utilized to shorten survey after dropping select items. Nonparametric correlation analysis (Spearman's rho) was conducted between the global sleep scores of the shortened and original surveys. Agreements tests (Kappa and McNemar's test) measured the agreement of the surveys and sensitivity and specificity were evaluated.

Results: Six factors were examined using maximum likelihood factoring method, applying squared multiple correlations with Promax rotation to allow for correlated variables. FA with six factors explained 100% of shared variance based on eigenvalues and accounted for 61% of variability based on variables. The FA resulted in 13 selected questions ("shortPSQI"), corresponding to 5 of the 7 components of the original survey. High correlation was found between the global scores of the original survey and the "shortPSQI" (rho = 0.94, p < 0.001). When the global score was converted to the categorical variable of good or poor sleepers, the agreement test indicated strong agreement (Kappa 0.83, 95% CI 0.79-0.86, p < 0.0001).

Conclusion: The validated, 19-item PSQI survey was shortened to 13 items. Tests of correlation and agreement indicate the "shortPSQI" may be an acceptable alternative to the original survey for young adults.

Clinical trial registration: Data for this study was taken from the Get Fruved study, registered on October 21, 2016, on clinicaltrials.gov (NCT02941497).

Figure 1

Graphical representation of seven clinically determined components (ovals) of original PSQI and their corresponding questions (rectangles) as well as the factor analysis-based six factors (triangles).

Figure 2

Spearman correlation of 0.94 (p < 0.0001) was detected between scores of original PSQI and shortened survey. Grey lines represent maximal possible global score in original PSQI (21) and “shortPSQI” (15). Blue vertical line represents cut-off 5 on PSQI; all who had more than 5 were assigned as poor sleepers. Red horizontal line represents cut-off point of 4 in shortPSQI; all students above the line were defined as poor sleepers.

Figure 3

Receiver Operating Characteristic (ROC) curve. The ROC curve is generated from seven different cut-off points of global scores on “shortPSQI” survey, using sensitivity (true positive) and 1 − specificity (specificity = true negative; 1 − specificity = false negative) with respect to cut-off 5 on original PSQI questionnaire. The cut-off of >4 on the global score of the “shortPSQI” demonstrates the highest sensitivity and specificity compared to the other cut-off values.