Program Schedule

1004
Technical and Logistical Issues in Incorporating Statistical Process Control into Healthcare-Associated Infection Surveillance Programs

Session: Poster Abstract Session: Surgical Site Infections
Friday, October 10, 2014
Room: The Pennsylvania Convention Center: IDExpo Hall BC
Background:

Statistical process control (SPC) can be an effective complement to healthcare-associated infection (HAI) surveillance. Since this use of SPC somewhat differs from its manufacturing origins, some issues warrant exploration to maximize SPC’s value in detecting outbreaks and reducing HAIs. Little empirical investigation has been conducted to develop a framework for managing signals from such a system.

Methods:

A team of HAI (DU) and SPC (NU) experts retrospectively applied conventional SPC methods to 10 years of de-identified surgical site infection (SSI) data from 40 community hospitals in the Duke Infection Control Outreach Network (DICON), focusing on 8 known SSI outbreaks. We met bi-weekly to review results and issues potentially important to implementing such an approach. Discussions were organized around 3 categories: detection performance and optimization, alternate methods, and logistics of managing resultant outbreak signals. Pilot investigations were conducted using empirical data and Monte Carlo analysis.

Results:

Shewhart and EWMA control charts detected all 8 outbreaks 0-12 months earlier than originally identified via standard surveillance. Table 1 illustrates the detection of one outbreak 0-3 months early. Issues affecting detection included what baseline data, aggregation level (hospital, surgeon, procedure), and EWMA smoothing factor to use. The value of dual, start-up, and sequential probability ratio SPC methods was identified as worthwhile to investigate. 

Conclusion:

While SPC methods appear useful for HAI surveillance, several technical considerations need addressing to maximize their benefit, each with potential impact on detection performance. Other SPC methods also may improve detection speed and therefore warrant both theoretic and empirical investigation.

Table 1: Empirical performance of Shewhart and EWMA charts illustrating one known outbreak, using different design parameters and baselines (A: Hospital’s year 1 data, B: DICON average benchmark).

Performance

Shewhart

EWMA (λ = 0.2)

EWMA (λ = 0.4)

A

B

A

B

A

B

# of months from first signal until known outbreak

-

3

-

2

-

3

# of signals in 12 months before known outbreak

0

3

0

3

0

3

# of total signals in non-outbreak years (before and after outbreak)

1

7

1

17

2

12

Arthur W. Baker, MD1, Deverick Anderson, MD, MPH1, Daniel J. Sexton, MD, FIDSA1, James Benneyan, PhD2, Salah Haridy2 and Nicholas Andrianas Jr.2, (1)Division of Infectious Diseases, Duke University Medical Center, Durham, NC, (2)Northeastern University Healthcare Systems Engineering Institute, Boston, MA

Disclosures:

A. W. Baker, None

D. Anderson, None

D. J. Sexton, None

J. Benneyan, PhD, None

S. Haridy, None

N. Andrianas Jr., None

Previous Abstract | Next Abstract >>

Findings in the abstracts are embargoed until 12:01 a.m. EDT, Oct. 8th with the exception of research findings presented at the IDWeek press conferences.

Sponsoring Societies:

© 2014, idweek.org. All Rights Reserved.

Follow IDWeek