Application of Data Mining to Intensive Care Unit Micro Data(1).

We describe refinements to and new experimental applications of the Data [ILLEGIBLE TEXT] Surveillance System (DMSS), which uses a large electronic health-care database monitoring emerging infections and antimicrobial resistance. For example, [ILLEGIBLE TEXT] from DMSS can indicate potentially important shifts in infection and antimicrobial resistance patterns in the intensive care units of a single health-care facility.

We have defined a new exploratory data mining process for automatically identif[ILLEGIBLE TEXT] unexpected, and potentially interesting patterns in hospital infection control and surveillance data. This process, and the system based on it, Data Mining [ILLEGIBLE TEXT] (DMSS), use association rules to represent outcomes and association role [ILLEGIBLE TEXT] monitor changes in the incidence of those outcomes over time. Through [ILLEGIBLE TEXT] infection control data from the University of Alabama at Birmingham Hospital, demonstrated that DMSS can identify potentially interesting and previously [ILLEGIBLE TEXT] Future work on prospective clinical studies to determine the usefulness of DMSS infection control is needed, as is improved event presentation for the user and [ILLEGIBLE TEXT] handling larger datasets.

The statistical strategies developed for automatically detecting temporal patterns surveillance data require that analysts explicitly define outcomes of interest [ILLEGIBLE TEXT] begins. The Data Mining Surveillance System (DMSS), on the other hand, is not to monitoring changes in user-defined outcomes. In DMSS, complex outcomes [ILLEGIBLE TEXT] by association roles, and outcome incidence is captured monthly.

An early version of DMSS, along with association roles and early experiments [ILLEGIBLE TEXT] organism, has been described (1). We briefly describe a newer version of DMSS experimental results obtained by using it to analyze 1 year's data from intensive [ILLEGIBLE TEXT] (ICUs) at the University of Alabama at Birmingham Hospital.

DMMS uses the following definitions. An itemset is a subset of the set of all [ILLEGIBLE TEXT] support of an itemset x, sup (x), is the number of records that contain x. If sup (x) where FSST is the frequent set support threshold (FSST), then x is a frequent set. association rule, A ?? B, where A and B are frequent sets and A n B = ??, is a [ILLEGIBLE TEXT] how often the items of B are found with the items of A. The incidence [ILLEGIBLE TEXT] denoted ip(A ?? B), is equal to sup(A ?? B)/sup(A). The precondition support of rule A ?? B is sup(A). The incidence proportion of an association rule A ?? B partition [p.sub.i] describes the incidence of the outcome, B, in the group, A, during [ILLEGIBLE TEXT] of incidence proportions for A ?? B from partitions [p.sub.1], [p.sub.2], ..., [p.sub.n] describes the the outcome B in group A from [t.sub.1] through [t.sub.n]. Therefore, by analyzing the series proportions of an association rule A ?? B, it should be possible to detect [ILLEGIBLE TEXT] trends in the incidence of B in A over time. In this way, surveillance of B in A is [ILLEGIBLE TEXT]

Bacterial susceptibility and related demographic data of patients in the [ILLEGIBLE TEXT] at Birmingham Hospital ICUs (medical, surgical [SICU], cardiac, neurologic [ILLEGIBLE TEXT] 1997 were extracted from the PathNet laboratory information system. Each [ILLEGIBLE TEXT] single isolate and contains the following data elements: date of admission, date [ILLEGIBLE TEXT] collection, date of results reported, source of isolate (e.g., sputum, blood), [ILLEGIBLE TEXT] organism Gram stain and morphologic features, patient's location in the hospital, (R), intermediate (I), or susceptible (S) test results to relevant antibiotics, [ILLEGIBLE TEXT] National Committee for Clinical Laboratory Standards MIC breakpoints (2).

Duplicate records were removed so that for each patient, no more than one [ILLEGIBLE TEXT] organism per month was included. In each remaining record, certain [ILLEGIBLE TEXT] were removed (only drugs to which the organism is historically susceptible at [ILLEGIBLE TEXT] time remained). Additionally, items of the form S~Antimicrobial were removed I~Antimicrobial and R-Antimicrobial items remained. Finally, data were [ILLEGIBLE TEXT] partitions ([p.sub.1] ... [p.sub.n]) before analysis. For each partition [p.sub.i], all frequent sets with [ILLEGIBLE TEXT] least 3 (FSST [is greater than] 2) and association rules with precondition support greater than 5 generated. Both the frequent set discovery and association rule-generating [ILLEGIBLE TEXT] beyond the scope of this review (3).

Each generated association rule must pass a set of rule templates that describe [ILLEGIBLE TEXT] interesting and uninteresting rules. Each template is a construct of the form [be.sub.1] = where [be.sub.1] and [be.sub.2] are Boolean expressions over items and attributes. [ILLEGIBLE TEXT] B satisfies rule template [be.sub.1] ?? [be.sub.2] if A satisfies [be.sub.1] and B satisfies [be.sub.2]. Two [ILLEGIBLE TEXT] association rule templates are used: include templates and exclude templates. [ILLEGIBLE TEXT] rule A ?? B passes a set of rule templates if A ?? B satisfies at least one [ILLEGIBLE TEXT] the set and does not satisfy any exclude template in the set.

Rule templates are handcrafted by domain experts to eliminate inherently [ILLEGIBLE TEXT] nonsense rules. This is accomplished through iterative experiments with [ILLEGIBLE TEXT] initially using few templates and then creating and modifying templates on the [ILLEGIBLE TEXT] review.

History is a database that holds association rules and their incidence proportions data partitions. In DMSS, the user specifies a set of rule templates that contains a inclusive and restrictive templates (Table 1). Only association rules that pass the are included in the history. To establish a baseline for an association rule, the [ILLEGIBLE TEXT] proportions of the role for the three previous partitions are obtained and stored [ILLEGIBLE TEXT] Once stored in the history, a rule is updated for each new partition regardless of it is generated in the partition. Therefore, for every association rule, the history [ILLEGIBLE TEXT] to-date time-series of incidence proportions.

Table 1. Templates used to filter association rules

Template   Left             Right
type       ([be.sub.1])     ([be.sub.2])     Explanation

Exclude    (R~Antibiotic)   (Anything)       Want antibiotic
                                             sensitivity
                                             [ILLEGIBLE TEXT]
                                             only.

Exclude    (Anything)       (Source)         Source of infection
                                             is not an
                                             [ILLEGIBLE TEXT]
                                             Therefore, exclude
                                             all rules with a
                                             [ILLEGIBLE TEXT]

Exclude    (NS OR Org       (NS OR Org       NS, Org, and GrMp
           GrMP)            OR GrMP          are more
                                             [ILLEGIBLE TEXT]
                                             kept together in
                                             either a
                                             [ILLEGIBLE TEXT]
                                             outcome.

Exclude    (Loc)            (Org OR GrMp)    If the left contains
                            AND              location, the rules
                            (R~Antibiotic)   that have Org and
                                             R~Antibiotic
                                             [ILLEGIBLE TEXT]
                                             R~Antibiotic.

Include    (Org OR Loc)     (R~Antibiotic    Include rules whose
                            OR               groups are specific
                            GrMp OR Org)     and whose outcomes
                            AND Not          are [ILLEGIBLE TEXT]
                            (Loc)            specific.

[be.sub.1] and [be.sub.2], Boolean expressions; R, resistant; NS, nosocomial; OR, "or"; Org, organism; GrMp and morphology; Loc, Location.

By analyzing information stored in the history, DMSS generates alerts that [ILLEGIBLE TEXT] extreme change in the incidence of an outcome B in a group A over time. For [ILLEGIBLE TEXT] 2 describes the incidence of Acinetobacter baumannii in a nosocomial tracheal [ILLEGIBLE TEXT] SICU isolates over the past six partitions. Clearly, a shift in incidence occurs [ILLEGIBLE TEXT] 4 months and the most recent 2 months of the series. If we call months 1, 2, 3, [ILLEGIBLE TEXT] window, [w.sub.p], and months 5 and 6 the current window, [w.sub.c], we can ask if there is a change in the incidence between [w.sub.p] and [w.sub.c]. We compute the cumulative [ILLEGIBLE TEXT] for [w.sub.p] (0/43) and for [w.sub.c] (5/18) and compare the two by a statistical test of two [ILLEGIBLE TEXT] generate an alert for an association rule r, DMSS first constructs a current [ILLEGIBLE TEXT] past window ([w.sub.p]) on the series of incidence proportions of r ([w.sub.c][r,0], [w.sub.p][r,0] [ILLEGIBLE TEXT] algorithm in the Figure). Second, it computes the cumulative incidence [ILLEGIBLE TEXT] window. Third, it compares the two cumulative incidence proportions by a test [ILLEGIBLE TEXT] proportions. Finally, if the difference between the proportions is statistically [ILLEGIBLE TEXT] = 0.01), it generates an alert. The value of [square] is user-defined and rather arbitrary. [ILLEGIBLE TEXT] not generated, the next set of current and past windows is formed ([w.sub.c][r, 1], [w.sub.p][r, [ILLEGIBLE TEXT] algorithm in the Figure), and the cumulative incidence proportions are compared. pairs are generated for the same association rule until an alert is generated or no pairs remain to be formed. DMSS generates all alerts by executing the procedure every association rule in the history.

Figure. Algorithm used to construct current and past windows for association rule r.

i=0; k=0
while (pc-2i-1 exists for r){
        j=0
        while (pc-2i-1-j exists for r){
                       i
        wc[r,k] = [union]pc-n
                n=0
                  i+j
        wp[r,k] = [union]pc-n - i - n - 1
                  n=0
          j++, k++
       }
        i++
}

Current and past window pairs are [ILLEGIBLE TEXT] algorithm in the Figure. If n is the [ILLEGIBLE TEXT] incidence proportions in the history for [ILLEGIBLE TEXT] ([w.sub.c]:[w.sub.p]) pairs are generated for that rule [ILLEGIBLE TEXT] following order: [MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII] For each pair, [w.sub.p] must be at least as [ILLEGIBLE TEXT]

The total number of events was reduced including all rules, to 36, by using the [ILLEGIBLE TEXT] Table 1; thus, classes of inherently [ILLEGIBLE TEXT] were eliminated. A retrospective look [ILLEGIBLE TEXT] events eliminated by the rule templates [ILLEGIBLE TEXT] they were uninformative. Therefore, the of templates resulted in a more focused DMSS output.

Of the 36 events, 18 were judged potentially interesting. Table 3 contains several representative events, one per row. Each row contains the association rule, the in proportions in [w.sub.c] (bold), and the incidence proportions in [w.sub.p] (nonbold). For [ILLEGIBLE TEXT] in Table 3 describes an increase in the number of Staphylococcus aureus [ILLEGIBLE TEXT] clindamycin, and erythromycin isolated from tracheal aspirates in the fourth [ILLEGIBLE TEXT] compared with those isolated in the 2nd and 3rd partitions. Of the events [ILLEGIBLE TEXT] only the NICU and SICU had events that were location-specific (Table 3), while were not.

Table 3. Representative events identified and considered of potential interest

                                                 Partition
Left                  Right
Denominator           Numerator           1      2      3      4

Staphylococcus   ??   R~Oxacillin(a,b)          0/10   0/8    7/14
aureus Source         R~Clindamycin
TRACHASP(c)           R~Erythromycin

NSNoso(d)        ??   R~Ceftazidime                           3/88

NP_GNR(e)        ??   R~Piperacilli                           0/17

NP_GNR           ??   R~Piperracillin                  1/12   0/14

NSNoso           ??   S. aureus          3/26   3/26   2/28   6/27
LocNICU(g)

                                             Partition

Left                  Right
Denominator           Numerator           5       6     7

Staphylococcus   ??   R~Oxacillin(a,b)
aureus Source         R~Clindamycin
TRACHASP(c)           R~Erythromycin

NSNoso(d)        ??   R~Ceftazidime      11/70

NP_GNR(e)        ??   R~Piperacilli       6/14

NP_GNR           ??   R~Piperracillin     4/11   4/8

NSNoso           ??   S. aureus           5/20   3/11
LocNICU(g)

Left                  Right
Denominator           Numerator          [ILLEGIBLE TEXT]

Staphylococcus   ??   R~Oxacillin(a,b)   [ILLEGIBLE TEXT]
aureus Source         R~Clindamycin
TRACHASP(c)           R~Erythromycin

NSNoso(d)        ??   R~Ceftazidime      [ILLEGIBLE TEXT]

NP_GNR(e)        ??   R~Piperacilli      [ILLEGIBLE TEXT]

NP_GNR           ??   R~Piperracillin    [ILLEGIBLE TEXT]

NSNoso           ??   S. aureus          [ILLEGIBLE TEXT]
LocNICU(g)

(a) R, resistant.

(b) Oxacillin, resistance implies resistance to amoxycillin/clavulanic acid, cephalothin, and cefazol

(c) Source TRACHASP, tracheal aspirates.

(d) NSNoso, nosocomial (3 days from admission).

(e) NP_GNR, non-pseudomonas gram-negative rod.

(f) LocSICU, location, surgical intensive care unit (SICU).

(g) LocNICU, location, neonatal intensive care unit (NICU).

The events identified by DMSS must be investigated by domain experts to [ILLEGIBLE TEXT] actual importance. In this example, the data burden was small since in a prospect only a few events would be presented to the user each month, thus allowing for [ILLEGIBLE TEXT] investigation of each event.

We believe that this approach to surveillance will allow hospital infection [ILLEGIBLE TEXT] focus their limited resources on issues of probable significance. We also believe approach is a step toward the public health surveillance system described by [ILLEGIBLE TEXT] Panter-Conner (4).

Table 2. A sample event generated by the Data Mining Surveillance System

                                    [p.sub.c-5]
Association rule                        (a)       [p.sub.c-4]

(nosocomial   ??   {Acinetobacter      0/11          0/10
SICU(b),             baumannii}
tracheal
aspirate

                                          [w.sub.p](c)

Association rule                    [p.sub.c-3]   [p.sub.c-2]

(nosocomial   ??   {Acinetobacter       0/9          0/13
SICU(b),             baumannii}
tracheal
aspirate

Association rule                    [p.sub.c-1]   [p.sub.c]

(nosocomial   ??   {Acinetobacter       2/9          3/
SICU(b),             baumannii}
tracheal
aspirate

                                             [w.sub.c]

(a) [P.sub.c], current pair.

(b) SICU, surgical intensive care unit.

(c) [w.sub.p], past window; [w.sub.c], current window.

(1) Presented in part at the International Conference on Emerging Infectious Diseases, March 8-11, Georgia.

References

(1.) Brossette SE, Sprague AP, Hardin JM, Waites KB, Jones WT, Moser SA. rules and data mining in hospital infection control and public health [ILLEGIBLE TEXT] Med Inform Assoc 1998;5:373-81.

(2.) National Committee for Clinical Laboratory Standards. Methods for [ILLEGIBLE TEXT] antimicrobial susceptibility tests for bacteria that grow aerobically. 4th ed. standard. NCCLS document M7-A4. Wayne (PA): The Committee; 1997.

(3.) Brossette SE. Data mining and epidemiologic surveillance [dissertation]. (AL): University of Alabama at Birmingham; 1998.

(4.) Dean AG, Fagan RF, Panter-Conner BJ. Computerizing public health [ILLEGIBLE TEXT] systems. In: Teutsch SM, Churchill RE, editors. Principles and practice of surveillance. New York: Oxford University Press; 1994. p. 200-17.

Comments/Responses

Have a comment on this article? Please use this form to reply. We're always happy to hear your views.

[ILLUSTRATION OMITTED]

Home | Top of Page | current Issue | Expedited | Upcoming Issue | Past Issue | EID Search | Contact Us

CDC Home | Search | Health Topics A-Z

This page last reviewed July 1, 1999

Emerging Infectious Diseases Journal National Center for Infectious Diseases Centers for Disease Control and Prevention

URL: http://www.cdc.gov/ncidod/eid/vol5no3/moser.htm

Stephen A. Moser, Warren T. Jones, and Stephen E. Brossette The University of Alabama at Birmingham, Birmingham, Alabama, USA

This work was supported in part by cooperative agreement U47-CCU411451 with the Centers [ILLEGIBLE TEXT] Control and Prevention (SAM) and a predoctoral research fellowship LM-00057 from the Nation Medicine (SEB).

Dr. Moser is associate professor, Department of Pathology, University of Alabama at [ILLEGIBLE TEXT] director of Laboratory Information Services, associate director of Clinical Microbiology for [ILLEGIBLE TEXT] and director of the Pathology Informatics Section. His research interests are applied research in [ILLEGIBLE TEXT] microbiology and the application of software as an aid to the intelligent analysis of medical [ILLEGIBLE TEXT] especially that generated in laboratory medicine.

Address for correspondence: Stephen A. Moser, University of Alabama at Birmingham, [ILLEGIBLE TEXT] Pathology, P246, 619 19th St., South Birmingham, AL 35233-7331, USA; fax: [phone]; [e-mail].