Missing Patterns May Determine Results and Limit Generalizability

• Activity Number: 557
• Type: Contributed
• Date/Time: Wednesday, August 3, 2016 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Consulting
• Author(s): Yumei Cao* and Joshua Field and Joel Linden and Pippa Simpson
• Companies: Medical College of Wisconsin and Blood Center of Wisconsin and La Jolla Institute for Allergy and Immunology and Medical College of Wisconsin
• Keywords: Missing data

Abstract

Introduction. The objective of the study was to compare invariant NKT (iNKT) cell activation at baseline and pain in 75 adults with sickle cell disease (SCD). The data was confounded by missing values. Methods. Based on the pattern of missing data, four groups were defined: complete (including baseline and pain values) data (n=19); baseline values only (n=32); pain values only (n=18); and all missing values (n=6), which were excluded from analysis. Paired and non-paired tests were performed to compare baseline, pain and baseline versus pain across and within groups. Additionally, demographic and clinical variables were compared among the groups. Results. Among those with complete data, there was no difference between baseline and pain iNKT cell activation level. However, iNKT cell activation in adults with pain was greater among the pain only group versus the complete group, likely because they were significantly different in treatment (hydroxyurea therapy) and disease severity (emergency department and hospital visits) and pain rating. Conclusion: The missing pattern reflected different types of patients limiting generalizability of results.