Data Sources. A retrospective cohort study of patients hospitalized in NYS was conducted. The Feinstein Institutes of Medical Research Institutional Review Board deemed that the study did not meet the definition of human subjects research; therefore, IRB review was not needed. Data were extracted from the New York Statewide Planning and Research Cooperative (SPARCS), and cross referenced by investigators with death data from Vital Statistics of New York State, to identify deaths that occurred after hospital discharge. All protocol methods and use of this data were carried out in accordance with relevant guidelines and regulations.
SPARCS is a comprehensive all payer data reporting system created in 1979 to collect information on hospital discharges from all Article 28 facilities operating in NYS. SPARCS is one of the largest data systems in the country, with over 2.3 million annual inpatient discharges and 6.7 million treat and release (i.e., same day) emergency room visits. (13) SPARCS currently collects patient level detail on patient characteristics, diagnoses and treatments, services, and charges for each hospital inpatient stay and outpatient visit. Death data were accessed from the New York State Vital Statistics Program, which registers live births, deaths, fetal deaths, induced terminations of pregnancy/abortions, etc.
Study Population and Selection Criteria. All patients aged ≥18 years hospitalized between January 1, 2014-December 31, 2016 were included in the analysis. A known CDI case was defined as one of ICD-9 and ICD-10 principal diagnostic codes: 00845, A0472, and A047. There were 3,714,486 total hospital discharges in NYS that met the eligibility criteria during the 2-year interval, of which 28,897 had a de novo CDI diagnosis (accounting for 0.78% of all discharges) and 3,685,589 did not. The total sample of 3,714,486 was a result of removing 23 hospital discharges (of which one was a CDI diagnosis) due to missing gender. Final propensity score matching (PSM) was 28,874 incident CDI cases successfully matched to 28,874 controls.
Exposure of Interest. The exposure of interest was diagnosis of CDI during hospitalization as documented in the medical chart. The data set was comprised of discharges between 01/01/14-12/31/16, the primary period of analysis corresponded to 07/01/2014-06/30/2016. There was a 6 month look back period (01/01/2014-06/30/2014) to determine whether a diagnosis of CDI in the analysis period was de novo (incident case) or a recurrent case. A 6 month follow up period (07/01/2016-12/31/2016) served to determine whether an exposed or control subject had any subsequent hospitalizations. A de novo CDI case was defined as a case of CDI in period of analysis, provided that there was no CDI visit for that patient in in the look back period, to avoid capture of recurrent infection. Controls were defined as all visits from subjects that never had a CDI diagnosis. Cases and controls who died in the hospital during the first visit were excluded from the analysis. In the event that a case or control died on the second or subsequent hospital visit, all data prior to that visit were included from the analysis.
Variables. Age, race, comorbidities (Charlson comorbidity index), insurance status, gender, ethnicity, 30,60, 90, 120, 180 days readmission, mortality within 7, 15, 30, 180 and 360 days of discharge, hospital LOS and the total charges.
Outcomes. The primary outcome in this study was 30-day readmission. Secondary outcomes included 60, 90, 120, 180 days readmissions, mortality within 7, 15, 30, 180 and 360 days of discharge, hospital LOS and total charges.
Propensity Score Matching (PSM). In this retrospective cohort study, comparability of the CDI (case) and non-CDI (control) groups regarding potential confounding variables. This was accomplished using PSM. Variables to compute the propensity score for each patient were: age, gender, Charlson Comorbidity Index (CCI), race, ethnicity, insurance type, and month/year of admission. PSM was accomplished using greedy nearest neighbor 1:1 matching based on logit (PS) with a caliper of 0.05, with an exact matching on month/year of admission.
Statistical Analyses. There were a total of 3,714,486 hospitalizations in New York State that met the eligibility criteria in the 2 year interval between 7/1/14 and 6/30/16, of which 28,897 were identified as having CDI. Of these, 28,874 incident CDI cases were successfully matched to 28,874 controls. Hospitalization that met eligibility critreria were identified and incident CDI cases determined. All cases, with the exception of the 23 instances of incomplete date, were successfully matched. The maximum standardized difference as percentage was 1.7% (Table 2), which is below the suggested upper limit of 10% (14, 15), and indicated a good balance between the matched groups. The variance ratios between the CDI and the pure control groups were between 1 and 1.03 for all variables in the matched observations, which is within the recommended range of 0.5 to 2.(16)
Table 2
Baseline Characteristics of the Incident CDI and Control groups in the Propensity Score Matched Sample
Variable
|
Incidence CDI cases
N=28,874
|
Controls
N=28,874
|
Standardized
difference in %s #
|
Variance Ratio
|
Age
|
67.55 ± 17.73 [median 70 (57-81)]
|
67.67±17.63 [median 70 (57-81)]
|
-0.626
|
1.01
|
Female sex
|
57.42%
|
57.34%
|
0.168
|
1.00
|
Race
|
|
|
|
|
White
|
68.94%
|
69.17%
|
-0.493
|
1.00
|
Black
|
13.92%
|
13.71%
|
0.573
|
1.01
|
Other
|
17.12%
|
17.14%
|
0.061
|
1.00
|
Ethnicity_class
Hispanic / Other
|
11.51%
|
11.38%
|
0.362
|
1.01
|
Insurance_status
|
|
|
|
|
Private / Other
|
20.62%
|
20.60%
|
0.056
|
1.00
|
Medicare
|
65.20%
|
65.47%
|
-0.557
|
1.00
|
Medicaid
|
14.18%
|
13.93%
|
0.618
|
1.01
|
Total CCI groups per record **
|
1.12±1.43 [median 0 (0-2)]
|
1.10±1.41 [median 0 (0-2)]
|
1.707
|
1.03
|
All categorical data are reported as percentages. The 2 continuous variables, Age and ‘Total CCI groups per record’ are reported as mean ±SD and median (25th,75th percentiles). |
Baseline characteristics of the incident CDI and control groups were compared using chi-square and two-sample t-tests, as appropriate (Table 1). Comparisons between the matched CDI and control groups were based on McNemar’s test (mortality and the readmission indicators), paired t-test (Log (total charges)) and Wilcoxon signed rank test (LOS). All categorical data are reported as percentages. Continuous variables are reported as mean ±SD and median (25th, 75th percentiles). Summary statistics for the outcomes, with the 95% confidence intervals (CI) for the difference between the matched groups are presented (Table 3). A result was considered statistically significant at the p<0.05 level. All analyses were performed using SAS version 9.4 (SAS Institute Inc, Cary, NC).
Table 1
Baseline Characteristics of the Incident CDI and Control groups
Variable
|
Incidence CDI cases
N=28,897
|
Controls
N=3,685,589
|
Standardized
difference in %s #
|
p-value*
|
Age
|
67.55±17.72 [median 70 (57-81)]
|
56.69±20.79 [median 58 (38-73)]
|
56.182
|
<.001
|
Female sex
|
57.41 %
|
57.59%
|
-0.365
|
0.5365
|
Race
|
|
|
|
<.001
|
White
|
68.91%
|
58.88%
|
21.012
|
|
Black
|
13.93%
|
18.65%
|
-12.833
|
|
Other
|
17.16%
|
22.47%
|
-13.345
|
|
Ethnicity_class Hispanic / Other
|
11.52%
|
16.18%
|
-13.499
|
<.001
|
Insurance_status
|
|
|
|
<.001
|
Private / Other
|
20.61%
|
31.58%
|
-25.184
|
|
Medicare
|
65.19%
|
41.95%
|
47.919
|
|
Medicaid
|
14.20%
|
26.46%
|
-30.842
|
|
Total CCI groups per record **
|
1.12±1.43 [median 0 (0-2)]
|
0.74±1.20 [median 0 (0-1)]
|
29.239
|
<.001
|
*P-values based on the Chi-square and the two-sample t-tests. |
# The standardized difference in% is the mean difference as a percentage of the average standard deviation. |
All categorical data are reported as percentages. The 2 continuous variables, Age and Total CCI groups per record are reported as mean±SD and median (25th, 75th percentiles). |
Table 3
Analysis of the Primary and the Secondary Outcomes
Outcome
|
Incidence CDI cases
|
Control
|
Diff (95% CI)
|
p-value*
|
30 d Readmission (n=
28,846 pairs)
|
28.26%
|
19.46%
|
8.81(8.11-9.5)
|
<.0001
|
60 d Readmission (n= 28,846 pairs)
|
37.65%
|
26.02%
|
11.62(10.87-12.38)
|
<.0001
|
90 d Readmission (n= 28,846 pairs)
|
42.93%
|
30.43%
|
12.5(11.72-13.28)
|
<.0001
|
120 d Readmission (n=28,846 pairs)
|
46.47%
|
33.74%
|
12.73(11.94-13.52)
|
<.0001
|
180 d Readmission (n=28,846 pairs)
|
51.39%
|
38.76%
|
12.63(11.82-13.43)
|
<.0001
|
Mortality within 7 days of discharge (n= 28,874 pairs)
|
3.68%
|
2.01%
|
1.67(1.4-1.94)
|
<.0001
|
Mortality within 15 days of discharge (n= 28,874 pairs)
|
5.76%
|
3.17%
|
2.59(2.26-2.93)
|
<.0001
|
Mortality within 30 days of discharge (n= 28,874 pairs)
|
8.65%
|
4.71%
|
3.93(3.53-4.34)
|
<.0001
|
Mortality within 180 days of discharge
(n= 28,874 pairs)
|
20.54%
|
11.96%
|
8.58(7.98-9.17)
|
<.0001
|
Mortality within 360 days of discharge
(n= 28,874 pairs)
|
26.15%
|
16.55%
|
9.6(8.93-10.26)
|
<.0001
|
LOS in days-Median (25-75th Perc.) (n= 28,874 pairs)
|
9 (5-16)
|
4 (2-7)
|
4 (0-11)
|
<.0001
|
Total Charges ($)- Geometric mean (95% CI)
|
58,430(57,689-59,181)
|
31,705(31,363-32,051)
|
1.84(1.81-1.87)
|
|
P-values were based on the McNemar’s test (Mortality and the Readmission indicators), the paired t-test (Log(Total Charges)) and the Wilcoxon signed rank test (LOS). |