Identifying Change Points and Forecasting Influenza Trends Using Diverse Influenza-like Illness Surveillance Data Capture Mechanisms in the City of Houston, Texas (2012-2016) Pubblico

Paul, Susannah (2017)

Permanent URL: https://etd.library.emory.edu/concern/etds/08612p322?locale=it
Published

Abstract

Background: Because influenza activity in a region is influenced by multiple factors, such as vaccine effectiveness, virus mutations, and travel, it is difficult to anticipate or identify significant activity increases early. Reports from traditional surveillance systems, which report data from emergency centers or providers, are accurate but delayed because of delays in patients seeing a provider or waiting for laboratory confirmation. Using novel surveillance methods for complementary information and combining available historical data can lead to earlier detection of influenza activity increases and decreases. Anticipating a surge would give public health professionals more time to prepare for a rise in cases and increase prevention efforts to reduce the risk of an epidemic.

Methods: Our objective was to investigate influenza activity in the City of Houston, by analyzing influenza-like illness data from diverse data capture mechanisms, from week 27 of 2012 through week 26 of 2016. Change point analysis was used to identify significant increase and decrease change points within each data source. ARIMA models were fitted for each source and used to estimate forecasts for influenza-like activity for the subsequent 10 weeks.

Results: All sources except for Flu Near You contained at least one significant increase and one significant decrease within each time interval. Overall, Athena, ILINet, and ER Centers resulted in similar start and end dates of the influenza season. Multiple, gradual changes within the typical influenza season and during non-seasonal time were identified. Forecasted estimates had wide confidence intervals with lower bounds below zero. The forecasted trend direction differed by data source, resulting in lack of consensus about future influenza activity.

Conclusion: The similarities in trend and timing of identified change points from outpatient information, ER Centers patient data, and ILINet influenza-like illness provider surveillance supports potential use of diverse data capture mechanisms to enhance influenza surveillance. Pooling the significant change points results in a comprehensive trend pattern identification for influenza activity. Though the forecasted estimates did not agree on trend direction and would be inaccurate for long-term predictions, pooling the predictions could be helpful in the short term if more historical information and influential variables are considered.

Table of Contents

Table of Contents

Introduction 1
Background and the Burden of Influenza 1
Influenza-like Illness and Syndromic Surveillance Systems 3
Gaps in Surveillance to Reduce Influenza Risk 6

Methods 10
Syndromic Surveillance Sources and Population 10
CPA Methods: Cumulative Sum and Bootstrapping 13
Imputing Missing Values 12
ARIMA Modeling and Forecasting 14
Stationarity and Differencing 15
Fitting an ARIMA Model 15
Obtaining Point Forecasts from ARIMA Models 16

Results 18
Identifying and Comparing Significant Change Points 18
Forecasting Future Influenza Activity Trends from ARIMA 26

Discussion 28
Strengths 28
Limitations 29
Further Considerations 31

Conclusion 32

Tables and Figures 34
Table 1. Peak Weeks per Year and ILI Percentages by Data Capture Mechanism 18
Figure 1. ILI Percentage by Data Capture Mechanism (2012-2016) 19
Figure 4. Significant Change Points by Data Capture Mechanism (2012 - 2016) 19
Table 4. Weeks (95% Confidence Interval) of Significant Change Points in ILI Percentages by Data Capture Mechanism (2012-2016) 24
Table 3. Peak Weeks per Year and ILI Counts by Data Capture Mechanism 34
Table 2. Stationarity and Seasonal Differencing Test Results 35
Table 5. ARIMA Model Orders and Parameters 35
Figure 3. ILI Counts by Data Capture Mechanism (2012-2016) 36
Figure 2. CUSUM Charts of ILI Percentages (Week 27 2012 - Week 26 2016) 37
Figure 5. Point Estimate Forecasts and 95% Confidence Intervals of ILI Percentages 43

References 46

About this Master's Thesis

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
School
Department
Degree
Submission
Language
  • English
Research Field
Parola chiave
Committee Chair / Thesis Advisor
Committee Members
Partnering Agencies
Ultima modifica

Primary PDF

Supplemental Files