Effect of Procalcitonin Testing on Health-care Utilization and Costs in Critically Ill Patients in the United States

Abstract

Background

There is a growing use of procalcitonin (PCT) to facilitate the diagnosis and management of severe sepsis. We investigated the impact of one to two PCT determinations on ICU day 1 on health-care utilization and cost in a large research database.

Methods

A retrospective, propensity score-matched multivariable analysis was performed on the Premier Healthcare Database for patients admitted to the ICU with one to two PCT evaluations on day 1 of ICU admission vs patients who did not have PCT testing.

Results

A total of 33,569 PCT-managed patients were compared with 98,543 propensity score-matched non-PCT patients. In multivariable regression analysis, PCT utilization was associated with significantly decreased total length of stay (11.6 days [95% CI, 11.4 to 11.7] vs 12.7 days [95% CI, 12.6 to 12.8]; 95% CI for difference, 1 to 1.3; P < .001) and ICU length of stay (5.1 days [95% CI, 5.1 to 5.2] vs 5.3 days [95% CI, 5.3 to 5.4]; 95% CI for difference, 0.1 to 0.3; P < .03), and lower hospital costs ($30,454 [95% CI, 29,968 to 31,033] vs $33,213 [95% CI, 32,964 to 33,556); 95% CI for difference, 2,159 to 3,321; P < .001). There was significantly less total antibiotic exposure (16.2 days [95% CI, 16.1 to 16.5] vs 16.9 days [95% CI, 16.8 to 17.1]; 95% CI for difference, –0.9 to 0.4; P = .006) in PCT-managed patients. Patients in the PCT group were more likely to be discharged to home (44.1% [95% CI, 43.7 to 44.6] vs 41.3% [95% CI, 41 to 41.6]; 95% CI for difference, 2.3 to 3.3; P = .006). Mortality was not different in an analysis including the 96% of patients who had an independent measure of mortality risk available (19.1% [95% CI, 18.7 to 19.4] vs 19.1% [95% CI, 18.9 to 19.3]; 95% CI for difference, –0.5 to 0.4; P = .93).

Conclusions

Use of PCT testing on the first day of ICU admission was associated with significantly lower hospital and ICU lengths of stay, as well as decreased total, ICU, and pharmacy cost of care. Further elucidation of clinical outcomes requires additional data.

FOR EDITORIAL COMMENT SEE PAGE 6

More than $20 billion was spent on sepsis care in 2011, making it the most expensive condition managed in US hospitals.¹ Sepsis is increasingly common, and despite advances in antimicrobial and supportive therapy, it continues to be associated with a high risk for morbidity and mortality.² The current consensus sepsis definition has been found to be both nonspecific and nonsensitive.³ Lack of a “gold standard” diagnostic test for sepsis has resulted in diagnostic dilemmas that may delay appropriate treatment and lead to poor outcomes.^{3, 4, 5} Unnecessary antibiotic use, as a consequence of sepsis overdiagnosis, is associated with increased length of stay (LOS), drug-related toxicities, Clostridium difficile infection, antimicrobial resistance, and health-care costs.⁶ Optimizing outcomes and minimizing resource utilization are contingent on timely and accurate discrimination between sepsis and mirroring conditions.

Procalcitonin (PCT), a precursor of calcitonin, has good specificity for a bacterial cause of suspected infection^{7, 8} and has been investigated for the diagnosis of bacterial infection, prognosis of sepsis, and management of antibiotic therapy.^{9, 10, 11, 12, 13, 14, 15} PCT guidance has been shown to reduce duration of antibiotic use in sepsis with no beneficial or detrimental impact on survival.¹⁰ Despite the emergence of PCT evaluation as a diagnostic criterion for sepsis in the 2012 Surviving Sepsis Campaign Guidelines, PCT testing has not been uniformly adopted, in part because of cost considerations.¹⁶ However, a Dutch study showed that PCT-based algorithms lowered hospital spending by €3503 per patient, saving the Netherlands health system an estimated €46 million annually.⁹

To provide current real-world information on health-care utilization and cost, we evaluated the impact of PCT testing performed on the first day of ICU admission in critically ill adult patients with suspected sepsis. An enhanced administrative database was used that included discharges from >550 US hospitals.

Patients and Methods

This study was a retrospective trial of patients aged ≥18 years admitted to an ICU with an admitting or discharge diagnosis code consistent with suspected or documented sepsis, septicemia, systemic inflammatory response syndrome, or shock (e-Table 1) between January 1, 2011, and May 31, 2014. Patients were grouped into those who did or did not receive 1 or 2 PCT determinations around the time of ICU admission. The initial PCT testing may occur in the ED or on the floor, but the PCT-tested patients were excluded if they received PCT determinations >2 days before or after ICU admission. The reason for PCT testing and the result were not specified in the administrative database that was used for the study. The database does not contain data to define the physician specialty who ordered the test or the patient characteristics that led to testing. The study design and methods were reviewed and approved by the Rush University Medical Center Office of Research Affairs per policy RA-IRB-118.

The data source was the Premier Healthcare Database, a large US hospital-based, service-level, all-payer administrative database. Utilization, cost, and outcome data were directly available in the database. Total antibiotic exposure was calculated as the total number of selected systemic antibiotics (e-Table 2) on a given day summed over the entire course of administration such that three antibiotics administered daily over 7 days would equal 21 days of exposure.

Because patients were not randomly assigned to PCT and non-PCT cohorts, differences in patient characteristics and severity may lead to bias in the estimated effect of PCT. To address this issue, 1:3 propensity score matching was implemented by using nearest neighbor matching to account for differences in the comparison groups.¹⁷ The analysis was limited to patients with a PCT determination within 1 day of ICU admission. Despite this approach, there were some control subjects with very short ICU stays who were discharged before the time that the PCT determination was run on the matched case. These control subjects were dropped because the shorter stays could not be attributable to a “treatment effect” (relating to whether a PCT determination was or was not obtained). Propensity scores were determined by using a logistic regression with receipt of PCT testing as the dependent variable and the following covariates: patient demographic characteristics (age categories, sex, race, ethnicity, health-care coverage type, and year of hospitalization), hospital characteristics (urbanicity, teaching status, bed size, and region), and patient’s clinical characteristics (admission type, admission source, number of types of antibiotics received on the first ICU day, dialysis on or before the first ICU day, ventilator use on or before the first ICU day, use of vasopressors or inotropes on the first ICU day, previous hospitalization within 30 days, and presence of the 10 most frequent admission diagnoses). After propensity score matching, the balancing properties of the matching covariates were examined through a review of the standardized differences between the PCT group and the non-PCT group. Standardized differences <0.1 were considered negligible.^{18, 19} Descriptive statistics of the outcomes were generated, and bivariate analysis was implemented on the propensity score-matched sample. The analytic and data processing plan are depicted in e-Figure 1.

Data measured on a continuous scale were summarized by using means; 95% CIs of the mean for the outcomes are also reported. Categorical data are expressed as counts and percentage of patients within each category. Bivariate analysis was performed to compare variables between those patients receiving PCT testing and those who did not. A t test was used to evaluate continuous variables, and χ² tests were used for categorical variables. P values <.05 were considered statistically significant.

Multivariable regression models were used to assess the association between PCT test and outcomes in the propensity score-matched sample. Control variables included the following: patient demographic characteristics (age categories, sex, race, ethnicity, health-care coverage type, and year of hospitalization), hospital characteristics (urbanicity, teaching status, bed size, and region), and patient’s clinical characteristics (admission type, admission source, number of different types of antibiotics received on the first ICU day, dialysis on or before the first ICU day, ventilator use on or before the first ICU day, use of vasopressors or inotropes on the first ICU day, previous hospitalization within 30 days, and presence of the 10 most frequent discharge diagnoses such as septicemia, other lung diseases, pneumonia with organism unspecified, heart failure, acute myocardial infarction, replacement and graft complication, other surgical complications that were not elsewhere classified, other bacterial pneumonia, acute renal failure, chronic bronchitis, pneumonitis due to solids and liquids, cardiac dysrhythmias, other forms of chronic ischemic heart disease, diabetes mellitus, and complications of medical care that were not elsewhere classified).

A supplemental analysis included the CareScience Mortality Risk Model score, a predictive value derived from clinical and demographic characteristics and available for those not transferred to another inpatient facility.²⁰ This score is calculated based on diagnosis codes and is available in the Premier Healthcare Database. Because we conducted a retrospective observational study using a hospital administrative database, the information needed to calculate The Acute Physiology and Chronic Health Evaluation Score (APACHE) or Simplified Acute Physiology Score (SAPS) (eg, vital signs, laboratory values) was not necessarily available, and these severity-of-illness scores could therefore not be determined. The missing values of the CareScience Mortality Risk Model score were due to the fact that some patients were transferred to other acute care facilities, and their mortality was not known. Because the basic assumption for imputation is that the missing values should be missing at random, the imputation may bring in bias for the group of patients transferred to another facility.

Generalized estimating equation models were used to model LOS, costs, and days of antibiotic therapy. These models with an exchangeable correlation structure generated robust variance estimates by adjusting for the within-hospital correlation. Logistic regression was used to model discharge status. The recycled prediction method²¹ was used to calculate the adjusted outcome values. Recycled prediction method is a way to calculate the predicted outcomes based on regression estimates. First, all patients were assumed to receive PCT, and the adjusted outcomes were predicted based on the regression coefficients, holding all covariates at their actual values. Second, predication was made assuming all patients were in the non-PCT cohort. The mean differences in the predicted values for the two hypothetical groups (PCT and non-PCT) were examined and the 95% CIs calculated by using the bootstrapping method. All statistical analyses were performed by using SAS version 9.2 (SAS Institute, Inc).

Results

This study evaluated 15,041,827 patients in the Premier Healthcare Database universe of patients; 730,088 had a potential sepsis, systemic inflammatory response syndrome, septicemia, or shock-related diagnosis on admission or discharge and were cared for in the ICU (Fig 1). This cohort was further stratified according to whether one or two PCT levels were obtained within the first day of ICU admission (n = 34,989) vs the non-PCT-managed cohort of ICU patients (n = 671,473). To reduce potential bias, a 1:3 propensity matching was performed. After matching, there were 98,543 non-PCT-managed patients at 570 hospitals and 33,569 PCT-managed patients at 286 hospitals. The data cleaning step to correct for the patients with extremely short stays affected 3% of control subjects. In most of these cases, the net effect was that <2% of cases had two or one rather than three matched control subjects. In 1% of case subjects, all three control subjects were affected; these cases were dropped. Even in this small number of affected patients, the numerical effect cannot be large because of the very short stays involved.

Figure 1.

Graphic illustration of the study flow from the Premier Healthcare Database universe of patients sorted according to ICU stay and selected sepsis, systemic inflammatory response syndrome, septicemia, and shock diagnoses and further sorted according to the use of PCT during the admission. Subsequent sorting used PCT testing on day 1 of ICU admission and a propensity score to match the study populations. PCT = procalcitonin.

There was no difference between the groups in the percentage of admissions from rural and urban hospitals, teaching and nonteaching hospitals, and bed size of the included hospitals (e-Table 3). The study populations were well matched for age, sex, race, ethnicity, health-care insurance status, and admission type (Table 1). The admission diagnosis, when present, was also similar between the two groups. No significant differences were noted in the number of systemic antibiotics administered to patients on the first day of their ICU stay or in the use of mechanical ventilatory support on or before the first day of ICU stay. There was no statistically significant difference in the use of vasopressors and/or inotropic medications on the first day of ICU stay for the two groups of patients.

Table 1.

Matched Patient Characteristics

Variable	Any PCT Test	No PCT Test	PCT vs No PCT, P Value	Standardized Differences
No. of discharges	33,569	98,543	…	…
Age, y			.424
Mean	66.29	66.37	…	0.005
SD	16.19	16.15	…	…

Variable	No. of Discharges (%)	No. of Discharges (%)	P Value	Standardized Differences
Sex			.666
Female	16,649 (49.6)	48,599 (49.3)	…	0.006
Male	16,917 (50.4)	49,934 (50.7)	…	0.006
Unknown	3 (0)	10 (0)	…	0.001
Race			.34
Black	3,813 (11.4)	11,048 (11.2)	…	0.005
Hispanic	40 (0.1)	130 (0.1)	…	0.004
Other	5,666 (16.9)	17,025 (17.3)	…	0.011
White	24,050 (71.6)	70,340 (71.4)	…	0.006
Ethnicity			.045
Hispanic	1,478 (4.4)	4,134 (4.2)	…	0.010
Non-Hispanic	28,452 (84.8)	83,334 (84.6)	…	0.005
Unknown	3,639 (10.8)	11,075 (11.2)	…	0.013
Health-care coverage type			> .999
Commercial	5,448 (16.2)	16,030 (16.3)	…	0.001
Medicaid	3,302 (9.8)	9,540 (9.7)	…	0.005
Medicare	22,027 (65.6)	64,976 (65.9)	…	0.007
Charity/indigent/self-pay	1,728 (5.1)	4,935 (5.0)	…	0.006
Other	1,064 (3.2)	3,062 (3.1)	…	0.004
Admission type
Elective	1,614 (4.8)	4,735 (4.8)	.996	0.000
Emergency	27,912 (83.1)	81,894 (83.1)	…	0.001
Other/unknown	27 (0.1)	81 (0.1)	…	0.001
Trauma center/urgent	4,016 (12.0)	11,833 (12.0)	…	0.001
Discharge status
Expired	6,469 (19.3)	17,948 (18.2)	< .001	0.027
Home	14,715 (43.8)	40,789 (41.4)	…	0.049
Other/unknown	336 (1.0)	935 (0.9)	…	0.005
Transferred to another health-care facility	12,049 (35.9)	38,871 (39.4)	…	0.073
Admission diagnosis
No admission diagnosis	5,965 (17.8)	16,508 (16.8)	< .001	0.027
Septicemia	4,432 (13.2)	12,859 (13.0)	.471	0.005
General symptoms	3,876 (11.5)	11,537 (11.7)	.427	0.005
Symptoms involving respiratory system and other chest symptoms	3,670 (10.9)	10,832 (11.0)	.763	0.002
Pneumonia, organism unspecified	2,680 (8.0)	8,071 (8.2)	.231	0.008
Other lung diseases	2,262 (6.7)	6,669 (6.8)	.854	0.001
Other symptoms involving abdomen and pelvis	771 (2.3)	2,307 (2.3)	.642	0.003
Cardiac dysrhythmias	552 (1.6)	1,552 (1.6)	.38	0.006
Heart failure	541 (1.6)	1,623 (1.6)	.659	0.003
Gastrointestinal hemorrhage	362 (1.1)	1,049 (1.1)	.831	0.001
Gastrointestinal system symptoms	414 (1.2)	1,220 (1.2)	.946	0.000
Certain adverse effects not elsewhere classified	491 (1.5)	1,373 (1.4)	.352	0.006
Hypotension	462 (1.4)	1,283 (1.3)	.303	0.006
Acute myocardial infarction	299 (0.9)	884 (0.9)	.915	0.001
Other urinary tract disorder	424 (1.3)	1,227 (1.2)	.798	0.002
No. of different types of antibiotics received on the first ICU day			.895
0	2,956 (8.8)	8,602 (8.7)	…	0.003
1	5,726 (17.1)	16,868 (17.1)	…	0.002
≥2	24,887 (74.1)	73,073 (74.2)	…	0.002
Any dialysis on or before ICU day 1	980 (2.9)	2,850 (2.9)	.797	0.002
Any ventilator use on or before ICU day 1	11,094 (33.0)	32,733 (33.2)	.571	0.004
Any use of vasopressors or inotropes on ICU day 1	10,425 (31.1)	30,461 (30.9)	.622	0.003

PCT = procalcitonin.

Figure 2 displays the success of the propensity score adjustment in the matched outcomes analysis without regression adjustment. In unadjusted analysis, PCT-guided care was associated with a significantly shorter hospital and ICU LOS (Table 2). PCT-guided care was also associated with significantly decreased total hospital, room and board, pharmacy, antibiotic, and laboratory costs. Antibiotic exposure was lower in the PCT-managed patients. More PCT-managed patients were discharged home, but inpatient mortality was slightly higher in this group. The PCT group was also less likely to be transferred to acute care, skilled nursing, intermediate care, or long-term care facilities.

Figure 2.

Graphic demonstration of assessment of the success of propensity score matching between the PCT-guided population and the non-PCT-guided study population. See Figure 1 legend for expansion of abbreviation.

Table 2.

Matched Outcomes Without Regression Adjustment

Variable	1 to 2 PCT Assessments Within 1 Day of ICU Admission (n = 33,569 discharges)		No PCT (n = 98,543 discharges)		Difference		PCT vs No PCT, P Value
	Mean	95% CI	Mean	95% CI	Mean	95% CI
LOS, d	10.5	10.4 to 10.6	12.8	12.7 to 12.9	−2.3	−2.4 to −2.2	< .001
ICU LOS, d	4.8	4.8 to 4.9	5.6	5.5 to 5.6	−0.7	−0.8 to −0.7	< .001
Total cost, $	25,513	25,163 to 25,864	33,164	32,898 to 33,429	−7,650	−8,090 to −7,211	< .001
ICU cost, $	16,814	16,560 to 17,069	21,630	21,429 to 21,832	−4,816	−5,140 to −4,492	< .001
Pharmacy cost, $	3,866	3,694 to 4,038	4,589	4,494 to 4,684	−723	−920 to −527	< .001
Antibiotic cost, $	832	768 to 897	936	911 to 961	−103	−173 to −34	< .001
Laboratory cost, $	1457	1,442 to 1,472	1,710	1,697 to 1,724	−253	−274 to −233	< .001
Room and board cost, $	12,212	12,053 to 12,370	15,379	15,254 to 15,503	−3,167	−3,369 to −2,965	< .001
Total antibiotic exposurea	14.9	14.7 to 15.0	17.3	17.2 to 17.4	−2.4	−2.6 to −2.3	< .001
Inpatient mortality	19.3%	18.8 to 19.7	18.2%	18.0 to 18.5	1.1%	0.6 to 1.5	< .001
Home	43.8%	43.3 to 44.4	41.4%	41.1 to 41.7	2.4%	1.8 to 3.1	< .001
Hospice	6.3%	6.1 to 6.6	6.4%	6.2 to 6.6	−0.1%	−0.4 to 0.2	.617
Other	1.7%	1.6 to 1.9	1.6%	1.5 to 1.7	0.1%	0 to 0.3	.122
Transfer to SNF/ICF/LTC	22.7%	22.2 to 23.1	25.2%	24.9 to 25.4	−2.5%	−3.0 to −2.0	< .001
Transfer to acute care	6.2%	5.9 to 6.4	7.2%	7.0 to 7.4	−1.0%	−1.3 to −0.7	< .001

ICF = intermediate nursing facility; LOS = length of stay; LTC = long-term care facility; SNF = skilled nursing facility. See Table 1 legend for expansion of other abbreviation.

^aTotal antibiotic exposure comprises both total number of systemic antibiotics administered and duration of administration.

Multivariable regression adjustment was performed to further address residual imbalances between patient and hospital characteristics. This analysis generally substantiated and refined the findings of the unadjusted matched analysis (Table 3). PCT-managed patients averaged 1.2 fewer days in the hospital (11.6 vs 12.7 days [95% CI for difference, 1.0 to 1.3]), 0.2 fewer ICU days (5.1 vs 5.2 days [95% CI for difference, 0.1 to 0.3]), and 0.7 day less of antibiotic exposure (16.2 vs 16.9 antibiotic days [95% CI for difference, 0.4 to 0.9]). In the PCT-managed group, total hospital costs were $2,759 less ($30,454 vs $33,213 [95% CI for difference, 2,156 to 3,321]), ICU costs were $1,310 less ($20,155 vs $21,465 [95% CI for difference, 847 to 1,702]), and pharmacy costs were $331 less ($4,238 vs $4,568 [95% CI for difference, 99 to 488]). Laboratory costs of the PCT-managed patients were $81 greater ($1,807 vs $1,726 [95% CI for difference, 51 to 114]). PCT-managed ICU patients were more commonly discharged home (44.1% vs 41.3% [95% for difference, 2.3 to 3.3]), but inpatient mortality was 0.7% greater (19% vs 18.3% [95% CI for difference, 0.3 to 1.2]).

Table 3.

Matched, Regression Adjusted Outcomes (N = 132,112)

Variable	PCT		No PCT		Difference		P Value
	Mean of Adjusted Value	95% CI	Mean of Adjusted Value	95% CI	Mean of Adjusted Value	95% CI
Total LOS	11.6	11.4 to 11.7	12.7	12.6 to 12.8	−1.2	−1.3 to −1.0	< .001
ICU LOS	5.1	5.1 to 5.2	5.3	5.3 to 5.4	−0.2	−0.3 to −0.1	.031
Total cost, $	30,454	29,968 to 31,033	33,213	32,964 to 33,556	−2,759	−3,321 to −2,156	< .001
ICU cost, $	20,155	20,625 to 19,798	21,465	21,270 to 21,710	−1,310	−1,702 to −847	< .001
Pharmacy cost, $	4,238	4,119 to 4,453	4,568	4,480 to 4,678	−331	−488 to −99	.002
Antibiotic cost, $	882	854 to 948	952	931 to 980	−70	−105 to 4	.074
Laboratory cost, $	1,807	1,778 to 1,839	1,726	1,710 to 1,744	81	51 to 114	.002
Total antibiotic exposurea	16.2	16.1 to 16.5	16.9	16.8 to 17.1	−0.7	−0.9 to −0.4	.006
Discharged to home	44.1%	43.7 to 44.6	41.3	41.0 to 41.6	2.8	2.3 to 3.3	.012
Discharged to hospice	6.3%	6.0 to 6.5	6.4	6.3 to 6.6	−0.2	−0.4 to 0.0	< .001
Discharged to other	1.8%	1.6 to 1.9	1.6	1.5 to 1.7	0.1	0.0 to 0.3	.779
Transfer to SNF/ICF/LTC	22.6%	22.2 to 23.0	25.2	24.9 to 25.5	−2.6	−3.1 to −2.2	< .001
Transfer to acute care	6.3%	6.0 to 6.5	7.2	7.0 to 7.3	−0.9	−1.2 to −0.6	< .001
Inpatient mortalityb	19.0%	18.6 to 19.4	18.3	18.0 to 18.5	0.7	0.3 to 1.2	.001
Inpatient mortalityc	19.1%	18.7 to 19.4	19.1	18.9 to 19.3	−0.0	−0.5 to 0.4	.93

95% CIs were calculated using bootstrap method with 500 replications. See Table 1 and 2 legends for expansion of abbreviations.

^aTotal antibiotic exposure comprises both total number of systemic antibiotics administered and duration of administration.

^bAll subjects.

^cAnalysis incorporating the CareScience Mortality Risk Model score; N = 127,809.

A sensitivity analysis limited to survivors was performed and yielded effectively identical results for utilization and cost as the all-patient analysis. An additional analysis added the CareScience Mortality Risk Model score (available for the 127,809 [96%] patients not transferred to another inpatient facility) to the previous multivariable regression. In this analysis, the observed mortality difference was no longer present (19.1% vs 19.1% [95% CI for difference, −0.5 to 0.4]).

Discussion

PCT testing on the first day of ICU care has received US Food and Drug Administration (FDA) clearance as a tool to assist in the identification of patients with a high likelihood of sepsis. A growing number of hospitals have adopted PCT testing to aid in the diagnosis and management of patients with potential sepsis. This study evaluated the large Premier Healthcare Database, which contained data from >619 million patient encounters (inpatient and outpatient) or approximately one in every five discharges in the nation over nearly 3.5 years to assess the impact of first ICU day PCT testing on health-care utilization and cost of care in adults with potential sepsis. Although this sample was not random, preliminary comparisons between patient and hospital characteristics for the hospitals that submit data to Premier and those of the probability sample of hospitals and patients selected for the National Hospital Discharge Survey suggest that the patient populations were similar regarding patient age, sex, LOS, mortality, primary discharge diagnosis, and primary procedure groups. The Premier Healthcare Database seems to reflect the landscape of US hospitals, which are primarily smaller nonteaching hospitals, but it also includes larger and academic medical centers, which comprise 5% and 7%, respectively, of the nation’s 5,627 hospitals in the 2016 American Hospital Association statistics.²² A comparison was made of the hospitals that comprise the Premier Hospital Database vs the American Hospital Association Database (e-Table 4); the results support the generalizability of the findings in this article because the study population has similarities to the overall composition of US hospitals.

To date, there is no approved “septic test” that identifies patients with sepsis with a sufficiently high likelihood to make it clinically useful.²³ PCT has been used, predominantly in Europe, to assist in identifying patients with respiratory infections who would benefit from antibiotic treatment.^{6, 8, 23, 24, 25, 26, 27} Studies have also suggested a potential benefit of using PCT testing to help with ICU prognosis and antibiotic stewardship related to decreasing antibiotic exposure.^{10, 24, 25, 26, 27, 28, 29} Neither of these uses has gained FDA approval.

The present study showed that the use of PCT testing on the first day of ICU care was associated with significantly lower hospital and ICU LOS. There was also a significant difference in the total hospital, ICU, and pharmacy costs when day 1 PCT testing in the ICU was used in critically ill adult patients. The PCT-managed patients had less total antibiotic exposure but higher laboratory costs. The all-patient analysis revealed a statistically significant, but only slightly higher (0.7%), risk of mortality that was not observed in an enhanced risk-adjusted analysis that included 96% of patients. This outcome is consistent with numerous prospective studies involving thousands of patients that have found no difference in mortality or other clinical outcomes using PCT guidance.^{24, 25, 26, 28, 30}

The interpretation of the latter findings is not readily apparent. One would certainly expect that more rapid identification and management of patients with sepsis would have also translated into improved survival in addition to improved health-care utilization and cost of care.^{16, 23, 31} Unfortunately, we cannot control for differences in severity of illness or chronic health conditions between the two cohorts, which may have affected survival. Similarly, we lacked information on subsequent use of vasopressors, ventilator support, renal replacement therapy, and other aspects of support that may influence outcome of the patient with sepsis.^{16, 23, 31, 32, 33, 34, 35} Another important factor that affects overall survival is the limitation and/or withdrawal of care, which can have a significant effect on overall outcome and does not necessarily reflect the adequacy of sepsis management but rather patient/family preferences or desires regarding end-of-life care.¹⁶

The present study is important in the validation of PCT testing’s ability to favorably affect the outcome of critically ill patients when used according to the FDA-cleared guideline. The study population was large and extremely diverse. The use of PCT was evaluated over nearly a 3.5-year period and in a variety of clinical settings, including academic and nonacademic institutions. The cost savings were real and consequential, exceeding the potential increased costs of laboratory testing associated with PCT testing on ICU admission.

The mechanisms resulting in the improved health-care utilization and cost of care are not immediately evident from this administrative database. It seems likely that PCT is being used both to help rule-out and rule-in sepsis on ICU admission. The former would directly save resources, whereas the latter would translate into earlier administration of appropriate antibiotic therapy and source control (ie, key components of sepsis management).^{16, 23} In addition, earlier recognition of the patient with sepsis may allow for earlier use of the appropriate resuscitation and management protocols (eg, the Surviving Sepsis Campaign Management Guideline), which have also been associated with improved outcome for patients with sepsis.^{16, 23, 31, 32, 33, 34, 35} Improved management using early effective appropriate antibiotics, source control, and fluid/vasopressor administration would be expected to lead to earlier patient improvement, reduced LOS, and decreased health-care utilization and cost.^{16, 31, 35} These outcomes were noted in this large patient population from a diverse group of hospitals and patient care settings.

The suspected timely and improved sepsis management in the PCT-managed patients was associated with decreased hospital, ICU, and pharmacy costs. These patients also required less antibiotic treatment as evidenced by decreased antibiotic exposure, which may have future implications on the ever-growing problem of bacterial resistance and development of multidrug-resistant organisms. Because patients with PCT testing after the first day of ICU care were not included in this database, it is unlikely that PCT was used to guide early antibiotic discontinuation. The higher laboratory costs (<$100) related to the use of PCT in the management of these critically ill patients may reflect the additional cost of PCT testing.

The major limitations of the present study include the retrospective nature of the analysis and the lack of clinical and certain other data. The identification of PCT testing was based on a text string search in the hospital billing records. Because hospitals recorded the cost of laboratory tests in various ways, we might not have captured all PCT tests in the hospitals. Second, as in all retrospective observational studies, the PCT and non-PCT cohorts were not randomly assigned. Therefore, there is a potential bias due to the difference in patient severity and unmeasured confounders. Although propensity score matching and multivariable analysis were implemented to control for the patient’s severity as much as possible, unobservable characteristics may still exist that were not controlled for in the analysis. The database lacked specific information as to whether there were hospital protocols governing the use of PCT in patient management or ICU admission. Similarly, there were no data regarding confirmatory culture results, the presence and site of actual infection, the adequacy of initial empiric antibiotic therapy, and/or source control measures, which may all affect outcome.^{16, 23, 32, 33, 34} There were also no data on subsequent need for vasopressor/inotropic therapy, ventilator support, renal replacement therapy, or the severity of illness, comorbidities, or limitations on aggressive care between the two groups, which could have affected treatment decisions and outcomes. All of these factors may influence health-care utilization and outcomes in severe sepsis and septic shock. Not all hospitals had onsite PCT testing available, and we do not have information protocols or decision processes that determined which patients would undergo PCT testing.

Conclusions

This evaluation of adult patients from US hospitals in the Premier Healthcare Database suggests that use of PCT testing on ICU admission was associated with a significant decrease in hospital and ICU LOS, less systemic antibiotic exposure, a slight increase in laboratory costs, and decreased hospital, ICU, and pharmacy costs. The significance and mechanisms surrounding the observed clinical outcomes warrant additional evaluation.