To determine the optimal number of fields counted, we measured the area of FF in 5, 10, and 20 fields in the same specimens two times. The %FF scores calculated were almost similar by 10- and 20-field measurement even when measuring at different times but were varied in 5-field measurement. Correlation between the scores at different times was significant in 10-and 20-field measurements (r = 0.970, p = 0.0365; r = 0.979, p = 0.0232, respectively) but not significant in 5-field measurement (r = 0.797, p = 0.2761). this
Thus, we measured at least 10 fields in each specimen. To compare our quantitative scoring method using the analytic software with semiquantitative scoring systems previously reported, each pathologist scored the profusion of FF semiquanti-tatively in three different ways, according to the Brompton, the Denver, and the Michigan FF scoring methods. When the score differed between the pathologists, a consensus score was reached. Interobserver variation was quantified using an unweighted к coefficient of agreement to take into account the degree of disagreement on the semiquantitative categorical scales. Clinical data, including sex, age, smoking history, symptoms, treatment, and outcome, were obtained from patient medical records. Laboratory findings, pulmonary function tests, and BAL data at the time of surgical lung biopsy were also recorded.
Statistical analysis was performed using statistical software (StatView J-4.5; SAS Institute; Cary, NC). Categorical data were compared between CVD-UIP and IPF/UIP using the x2 test for independence, and continuous data were compared using Mann-Whitney U test. The intraobserver and interobserver correlation, and the relationship between the quantitative %FF score and clinical continuous data were analyzed using the Pearson correlation coefficient. The relationship between the quantitative %FF score and the semiquantitative scores was analyzed using the Spearman rank correlation coefficient. The overall survival experience for each group was estimated using Kaplan-Meier curves. The log-rank test was used to compare survival between two groups. The effect of the quantitative %FF score on the risk of death was modeled using the Cox proportional hazards regression. All tests were two sided and performed at the 0.05 significance level.