Association of Unit-Level Telecritical Care Intensity of Service and Length of Stay in the Intensive Care Unit

Abstract

Background:

Telecritical care (TCC) has been shown to improve outcomes in the intensive care unit (ICU). A TCC was developed and implemented a nocturnal TCC across 10 ICUs in our Health System. TCC coverage patterns and level of involvement vary among ICUs. We identified an opportunity to determine the impact of TCC involvement on the ICU length of stay (LOS).

Objective:

The primary objective of this study was to assess if intensity of service provided by TCC impacts ICU LOS.

Methods:

This retrospective review was conducted for all patients admitted to covered ICUs during a 2-year period. ICUs were stratified by the coverage model provided by the TCC and the count of orders placed by the TCC served as a surrogate for intensity of service. Confounding variables were abstracted from the Acute Physiology and Chronic Health Evaluation (APACHE) databases. Spearman's rank correlation coefficient was used to measure the strength of the relationship between ICU LOS and TCC order volume. A linear regression model was used to describe the relationship between order volume and ICU LOS, while adjusting for confounding variables.

Results:

There is a strong negative relationship between TCC order volume and ICU LOS, as shown by the Spearman rank correlation coefficient of −0.818. The associated p-value of 0.0038 supports the strength of this relationship.

Conclusion:

Our results demonstrate the impact of nocturnal TCC involvement in patient care. As TCC order volume per ICU admission increases, ICU LOS decreases. We interpret this as an indication for deeper involvement between the TCC team and any on-site providers.

Introduction

Telecritical care (TCC) has been shown to improve clinical and operational outcomes in the intensive care unit (ICU).^1
–3 Yet, wider adoption of TCC is limited by cost, and persisting concerns about variable clinical, operational, and cost-effectiveness outcomes.^4

–11 These concerns spurred the elaboration of a research agenda for TCC to inform further development.¹² It called for research into mechanisms of effectiveness of TCC. Researchers have begun to address these questions.^13

–18 One study suggested that selectively focusing on high-risk patients will enhance cost-effectiveness.¹³ Ethnographic studies of TCC identified organizational characteristics as one domain of factors that affect program effectiveness. Other factors identified include staffing models, and degree of allowed involvement of the TCC team in patient care activities.

Our institution has implemented a nocturnal 7 p.m. to 7 a.m. TCC service (eHospital) across the health system. At some of the monitored facilities, there was either a resident physician or an advanced practice nurse (APN) in-house overnight. In those settings, the in-house providers maintain first-in-line responsibility for patient care, coordinating with the TCC team as needed. The other facilities had no in-house providers dedicated to the ICU overnight. So, the bedside nurse works primarily with the TCC team who has the first responsibility for patient care. On-call physicians coming back in when needed, for hands-on interventions.

Given the different in-house staffing models of the ICUs monitored by our TCC service, the number of patient care activities exclusively or predominantly delivered by the TCC team varies by site. Furthermore, the degree of coordination between the in-house and TCC teams also varies. As such, we postulate that variations in the degree of involvement of the TCC service could affect its impact on relevant outcomes such as length of stay (LOS) in the different ICUs. Accordingly, the primary objective of this study was to assess if the intensity of service from the TCC is associated with differences in ICU LOS.

Methods

This study was approved by the Cleveland Clinic Institutional Review Board with a waiver of informed consent as a retrospective review of prospectively collected unit-level data. We abstracted data from the electronic health records (EHR) datamart and the enterprise Acute Physiology and Chronic Health Evaluation (APACHE) registry. The data covered the period from January 1, 2017, to December 31, 2018. We used the APACHE data registry to identify all the patients whose admission and discharge to and from the ICUs monitored by eHospital fell within those dates.

STUDY SETTING

Cleveland Clinic's TCC service was sequentially deployed from 2014 to 2017 to 10 ICUs in 8 different hospitals within the health system. The TCC team (intensivists, advanced practice registered nurses, and registered nurses) uses a locally developed proprietary platform for patient triage. The platform is integrated with the system's EHR, into which the TCC team directly enters orders and notes. Three of the monitored hospitals are tertiary level with residency programs, and the rest are community hospitals, one of which also has some residency programs. Consequently, there were three models of nocturnal ICU coverage as noted earlier. Model 1 was eHospital only. Model 2 was eHospital with bedside APN, and model 3 was eHospital with bedside residents and/or intensivists.

Regarding authorization to provide care, the TCC team offers two levels of service. Level 1 covers emergencies such as cardiac arrest and, by default, is available to every patient. The TCC is empowered to get involved and provide needed care, akin to an in-house code-blue response team. Level 2 service allows the TCC team to manage every aspect of the patient's care—including changing previously determined plans of care, as necessary. Level 2 service is an opt-in option selected by the primary attending physician upon patient admission to the ICU.

Study Parameters

Critical care evaluation services encompass all diagnostic, monitoring, and therapeutic care administered to patients. The spectrum can include obtaining, reviewing, and documenting the following. A clinical history, physical examination, laboratory tests, diagnostic imaging, and hemodynamic data. All these, along with discussions with bedside nurses, ancillary staff, family members, and periodic surveillance rounding, also apply to the TCC setting, with the exception of hands-on physical examination or procedures. However, it is not easy to objectively measure the intensity of most of these services in terms of quantity, effort, and time.

However, every clinical intervention requires an order from a provider. With electronic order entry, the count of orders placed by the TCC team is easily determined and could serve as a surrogate for intensity of service. Clearly, this approach underestimates the clinical effort by the TCC providers, because not every care activity necessarily results in a new order. Given the different nocturnal bedside staffing models in the monitored ICUs, the order count is objective and allows easy delineation of services initiated by the TCC from those implemented by any in-house bedside providers. Accordingly, the average count of orders per admission at the unit level was used in the analysis as a measure of intensity of service.

The other relevant parameters abstracted for the study were the model of ICU coverage (types 1, 2, or 3 as described above) and ICU LOS. Potential confounding variables abstracted for the study included gender, age, race, APACHE 3 scores, APACHE active treatment, and mechanical ventilation status on day 1, as well as some specific conditions including elective or emergent surgery, sepsis, acute myocardial infarction, end-stage renal disease, and diabetes mellitus.

Statistical Analysis

Continuous variables were summarized with median [Q1, Q3], and categorical variables were summarized with frequency (%). Spearman's rank correlation coefficient was used to measure the strength of the relationship between ICU LOS and TCC order volume. A log-transformed linear regression model was used to further describe the relationship between order volume and ICU LOS, while adjusting for possible confounding variables.

ICU LOS was heavily skewed to the right, with a few high outliers. Accordingly, a log transformation was used to make the distribution more appropriate for linear regression. The results displayed take this transformation into account (the regression estimates are exponentiated) and may be interpreted as presented. The analysis was done in SAS software (version 9.4; Cary, NC) and p-values under 0.05 were considered statistically significant.

Results

During the study period, there were 8,347 admissions to the covered ICUs, representing 7,457 unique patients who had at least one order placed by the eHospital TCC service during their ICU admission. This is shown in Table 1, which also presents a summary of the data variables in the study. In total, the TTC team placed 116,895 patient care orders during the study period. Table 2 presents the admission volumes, average TCC order volume per admission, and LOS per admission, for each ICU. There is a strong negative relationship between TCC order volume and ICU LOS, as shown by the Spearman rank correlation coefficient of −0.818. The associated p-value of 0.0038 supports the strength of this relationship. As the TCC order volume per ICU admission increases, the ICU LOS decreases.

Table 1.

Data Summary

FACTORS	TOTAL (n = 8,350)
Age, median [Q1, Q3]	69.0 [57.0, 79.0]
Gender, No. (%)
Female	4,282 (51.3)
Male	4,068 (48.7)
Race, No. (%)
White/Caucasian	6,225 (74.6)
Black/African/Haitian	1,794 (21.5)
Other	331 (4.0)
Coverage model, No. (%)
Type 1 (TCC only)	5,414 (64.8)
Type 2 (nurse practitioner + TCC)	395 (4.7)
Type 3 (resident + TCC)	2,541 (30.4)
ICU classification, No. (%)
Community	4,621 (55.3)
Tertiary	3,729 (44.7)
APACHE III score,^a median [Q1, Q3]	56.0 [41.0, 74.0]
Active treatment on day 1, No. (%)	4,992 (59.8)
Mechanical ventilation on day 1, No. (%)	3,391 (40.6)
Acute myocardial infarction, No. (%)	39 (0.47)
End-stage renal disease, No. (%)	594 (7.1)
Diabetes mellitus, No. (%)	2,847 (34.1)
Sepsis, No. (%)	1,862 (22.3)
Surgical status, No. (%)
No surgery	7,701 (92.2)
Elective	532 (6.4)
Emergent	117 (1.4)
ICU length of stay, median [Q1, Q3]	2.3 [1.3, 4.5]
TCC order volume per ICU admission, median [Q1, Q3]	5.0 [3.0, 19.0]

Data not available for all subjects. Missing values: APACHE III score = 12.

APACHE, Acute Physiology and Chronic Health Evaluation; ICU, intensive care unit; TCC, telecritical care.

Table 2.

Intensive Care Unit Admission Volume, Average Order Volume, and Length of Stay

ADMITTING ICU	ICU ADMISSION VOLUME n (%)	AVERAGE ORDER VOLUME PER ADMISSION	AVERAGE LOS PER ADMISSION	SPEARMAN CORRELATION	p-VALUE
ICU #1	1,838 (22)	24.18	2.66	ρ = −0.818	H₀: ρ = 0 p-value = 0.0038
ICU #2	1,878 (22.5)	19.78	3.11
ICU #3	1,698 (20.3)	13.61	3.68
ICU #4	510 (6.1)	4.70	5.89
ICU #5	1,413 (16.9)	4.40	3.40
ICU #6	256 (3.1)	4.01	5.12
ICU #7	179 (2.1)	3.88	4.84
ICU #8	216 (2.6)	3.72	4.68
ICU #9	103 (1.2)	3.15	13.13
ICU #10	259 (3.1)	2.83	10.66

LOS, length of stay.

The distribution of average order volumes showed a clear demarcation of the monitored ICUs into the two groups (Table 2): Three ICUs had average order volumes over 13 per admission, whereas the other seven had average order volumes under 5. This clear dichotomy allowed us to select the value 10, as an arbitrary threshold. ICUs with average order volume over 10 were labeled as “High-Volume Units” and those under 10 as “Low-Volume Units.” Table 3 shows the relationship between coverage models at the monitored ICUs and high or low TTC order volumes. Admissions with high order volumes were in type 1 (TCC only) ICUs, while admissions with low order volumes were in types 2 and 3 ICUs. Subsequently therefore, we used order volume classification as an indicator for the coverage model in the regression analysis.

Table 3.

Intensive Care Unit Admission Volume by Coverage Model

ADMISSION VOLUME	ICU COVERAGE MODEL
ADMISSION VOLUME	TYPE 1 (TCC ONLY)	TYPE 2 (NURSE PRACTITIONER + TCC)	TYPE 3 (RESIDENT + TCC)	TOTAL
High-volume ICU	5,414	0	0	5,414
Low-volume ICU	0	395	2,541	2,936
Total	5,414	395	2,541	8,350

Table 4 presents the results for the log linear regression model. The model showed that being a “high order volume ICU” is a significant predictor of ICU LOS. The effect estimate is 0.739 (95% confidence interval = [0.708–0.771], p-value <0.001), meaning that a patient in a high-volume ICU unit is likely to have an LOS about 73.9% of a patient in a “low order volume ICU.” The model included other variables that were used to adjust for potential confounding. Several of these showed statistically significant relationship with ICU LOS, including ICU complexity, gender, APACHE score, and APACHE active treatment status on day 1.

Table 4.

Log Linear Regression Model, Modeling Intensive Care Unit Length of Stay

FACTOR	ESTIMATE	95% CONFIDENCE INTERVAL	p-VALUE
Intercept	3.012	2.188–4.148	<0.001
High order volume ICU unit (vs. low volume)	0.739	0.708–0.771	<0.001
ICU complexity: community (vs. tertiary)	0.947	0.908–0.987	0.011
Age	0.999	0.998–1.001	0.409
Gender: female (vs. male)	0.948	0.913–0.984	0.005
Race
Black/African/Haitian (vs. white/Caucasian)	1.024	0.977–1.074	0.320
Other (vs. white/Caucasian)	0.932	0.846–1.026	0.151
APACHE III score	1.005	1.005–1.006	<0.001
Active treatment on day 1: no (vs. yes)	0.641	0.615–0.669	<0.001
Acute myocardial infarction: no (vs. yes)	1.116	0.848–1.470	0.433
End-stage renal disease: no (vs. yes)	1.164	1.080–1.255	<0.001
Diabetes mellitus: no (vs. yes)	0.966	0.927–1.006	0.092
Sepsis: no (vs. yes)	0.804	0.758–0.853	<0.001
Surgical status
Elective (vs. no surgery)	1.136	1.046–1.235	0.003
Emergent (vs. no surgery)	1.092	0.926–1.287	0.294

Measurements which met the level of statistical significance have been highlighted in bold text.

Discussion

In our setting, there is a strong negative relationship between average TCC order volume and ICU LOS, as shown by the results of the Spearman rank correlation and supported by the results of the multivariable log linear model. Furthermore, several variables included in the model showed a statistically significant relationship with ICU LOS, but most of these may not have practical significance. Yet, the direction of some of these relationships (sepsis, end-stage renal disease, and APACHE active treatment status on day 1) is concordant with what would be expected for ICU LOS. Nonetheless, adjusting for these variables in the model further supports the inverse relationship between TCC order volume and ICU average LOS.

In practical terms, if a patient in a low order volume unit has a 24-h stay, a similar patient in a high order volume unit is expected to have about an 18-h stay. A reduction in LOS measured in hours might seem inconsequential. However, the downstream benefits of this on hospital operations and throughput are real. An open ICU bed allows for quicker admission of patients from the emergency room, operating room, or the regular ward, and may preclude the need to transfer patients to other facilities. Also, with flexibility in nurse staffing, adjustments can be made in the number of nurses assigned to the current or upcoming shifts. This is significant because labor accounts for the majority of ICU costs.

In our context, the distribution of TCC intensity of service, as measured by average volume of orders per admission, corresponded exactly with different models of bedside staffing in the monitored ICUs (Table 3). This indicates that the differential impact on ICU LOS is dependent on the bedside staffing model. Numerous reports have shown that TCC programs may have a variable impact in different ICUs.^9,19
–21 Our results suggest one mechanism that could explain this variability. Indeed, our results are consistent with studies that identified organizational characteristics as a factor for TCC program effectiveness.^16,22 Certainly, other TCC programs may not have such a clear correspondence between intensity of service and the bedside model of care. Nonetheless, our findings bear noting, insofar as any variation in TCC intensity of service is driven by the bedside model of care.

It may not be surprising to expect better outcomes when intensivists are primarily responsible for care, compared with residents and APNs. Yet, our TCC teams had placed some orders for patients in our models 2 and 3 ICUs. Clearly, they had some level of involvement, perhaps limited, in the care of those patients. As has been previously characterized, TCC and bedside teams can collaborate on urgent or routine ICU care, or for clinical decision-making and support.¹⁸ However, our retrospective review cannot capture the frequency and nature of the interactions between the TCC team and the on-site providers and nurses.

Mechanistically our findings could imply that the TCC should have first responsibility for patient care. This is too simplistic an interpretation and could be misleadingly stretched to mean that a bedside team is redundant. Rather, we think the degree and nature of involvement in the care are the more relevant variables. The coordination between the in-house and TCC teams can vary depending on several factors. Among others, the types of patients, severity of illness, degree of TCC program acceptance, level of experience of clinical and bedside teams, individual preferences, and nature of interpersonal interactions all play a role. The emphasis should be on improving collaboration between the TCC team and the on-site providers. Consequently, we view our results as highlighting the importance of optimal collaboration between TCC and bedside nurse practitioners and residents.

Along those lines, it is notable that our TCC team of intensivists, nurse practitioners, and nurses work side by side in the operations center. In this close setting, they maintain ongoing and informal collaboration as they simultaneously monitor and care for patients. This process of jointly contributing to patient care may be a crucial element. If so, it may be worthwhile to explore how to mimic such interactions between the TCC team and on-site bedside providers. Furthermore, our study covers a period that overlapped the last year of sequential program deployment (2017) and the first full year of TCC operations with no new sites added (2018). Thus, the TCC had 3.5 years of experience working together, whereas some of the types 2 and 3 ICUs had much less time and experience collaborating with the TCC team. It may be that outcomes would be similar in all settings, with a sufficient “wash-in” period after site activation.

Our study has all the usual limitations of retrospective reviews such as completeness of data and impact of uncollected variables. Moreover, our study is centered on a nocturnal TCC program within one health system, and this may limit the generalizability of results. Furthermore, we presume that a higher volume of TCC orders per admission indicates greater intensity of service in that patient's care. Yet, this surrogate measure for degree of TCC involvement does not capture every action by the TCC. This underscores the need to develop clear and unambiguous measures of TCC activity and utilization for use in assessing program effectiveness and the mechanisms thereof.¹⁸ Nonetheless, this underestimate is a negative bias against the TCC, thus strengthens our results. Furthermore, our TCC program had several years of experience before the study period, thus suggesting that our findings are robust. In all, our study adds to the accumulating insights into the mechanisms of effectiveness of TCC programs.

Conclusion

We have demonstrated an association between intensity of TCC service, as measured by volume of orders, and ICU LOS. In our context, this corresponded with a model in which our nocturnal TCC team has first-in-line responsibility for patient care. We interpret this as an indication for deeper involvement between the TCC team and any on-site providers. As such we call for more research to characterize the ideal workflows to optimize collaboration between the TCC team and on-site bedside providers.

Footnotes

Authors' Contributions

All authors have contributed to all parts of the research, including conceptualization, data analysis, and article preparation and review.

Acknowledgments

The authors wish to recognize with gracious appreciation the dedicated physicians, nurse practitioners, and nurses who deliver expert care nightly in the telecritical care operations center.

Disclosure Statement

No competing financial interests exist.

Funding Information

No funding was received for this article.

References

Lilly

, Zubrow

, Kempner

, Reynolds

, Subramanian

, Eriksson

, et al. Critical care telemedicine: Evolution and state of the art. Crit Care Med, 2014; 42:2429–2436.

Mackintosh

, Terblanche

, Maharaj

, Xyrichis

, Franklin

, Keddie

, et al. Telemedicine with clinical decision support for critical care: A systematic review. Syst Rev, 2016; 5:176.

Chen

, Sun

, Yang

, Liu

, Zhang

, Peng

, et al. Clinical and economic outcomes of telemedicine programs in the intensive care unit: A systematic review and meta-analysis. J Intensive Care Med, 2017; 33:383–393.

Young

, Chan

, Lu

, Nallamothu

, Sasson

, Cram

. Impact of telemedicine intensive care unit coverage on patient outcomes: A systematic review and meta-analysis. Arch Intern Med, 2011; 171:498–506.

Yoo

, Kim

, Sasaki

, Melnikow

, Marcin

. Economic evaluation of telemedicine for patients in ICUs. Crit Care Med, 2016; 44:265–274.

Udeh

, Udeh

, Rahman

, Canfield

, Campbell

, Hata

. Telemedicine/virtual ICU: Where are we and where are we going?. Methodist Debakey Cardiovasc J, 2018; 14:126–133.

Franzini

, Thomas

. Costs and effectiveness of tele-ICUs in reducing morbidity and mortality in intensive care units. J Med Econ, 2008; 11:165–169.

Breslow

, Rosenfeld

, Doerfler

, Burke

, Yates

, Stone

, et al. Effect of a multiple-site intensive care unit telemedicine program on clinical and economic outcomes: An alternative paradigm for intensivist staffing. Crit Care Med, 2004; 32:31–38.

Kahn

, Le

, Barnato

, Hravnak

, Kuza

, Pike

, et al. ICU telemedicine and critical care mortality: A national effectiveness study. Med Care, 2016; 54:319–325.

10.

Groves

Jr. , Holcomb

Jr ., Smith

ML.

Intensive care telemedicine: Evaluating a model for proactive remote monitoring and intervention in the critical care setting. Stud Health Technol Inform, 2008; 131:131–146.

11.

Pannu

, Sanghavi

, Sheley

, Schroeder

, Kashyap

, Marquez

, et al. Impact of telemedicine monitoring of community ICUs on interhospital transfers. Crit Care Med, 2017; 45:1344–1351.

12.

Kahn

, Hill

, Lilly

, Angus

, Jacobi

, Rubenfeld

, et al. The research agenda in ICU telemedicine: A statement from the Critical Care Societies Collaborative. Chest, 2011; 140:230–238.

13.

Yoo

, Kim

, Sasaki

, Hoch

, Marcin

. Selected use of telemedicine in intensive care units based on severity of illness improves cost-effectiveness. Telemed J E Health, 2018; 24:21–36.

14.

Shahpori

, Hebert

, Kushniruk

, Zuege

. Telemedicine in the intensive care unit environment—a survey of the attitudes and perspectives of critical care clinicians. J Crit Care, 2011; 26:328.e9–e15.

15.

Young

, Chan

, Cram

. Staff acceptance of tele-ICU coverage: A systematic review. Chest, 2011; 139:279–288.

16.

Kahn

, Rak

, Kuza

, Ashcraft

, Barnato

, Fleck

, et al. Determinants of intensive care unit telemedicine effectiveness. An ethnographic study. Am J Respir Crit Care Med , 2019; 199:970–979.

17.

Rak

, Kuza

, Ashcraft

, Morrison

, Angus

, Barnato

, et al. Identifying strategies for effective telemedicine use in intensive care units: The ConnECCT study protocol. Int J Qual Methods, 2017; 16 [Epub ahead of print]; DOI: 10.1177/1609406917733387.

18.

Thomas

, Moeckli

, Mengeling

, Goedken

, Bunch

, Cram

, et al. Bedside critical care staff use of intensive care unit telemedicine: Comparisons by intensive care unit complexity. Telemed J E Health, 2017; 23:718–725.

19.

Becker

, Fusaro

, Scurlock

. Telemedicine in the ICU: Clinical outcomes, economic aspects, and trainee education. Curr Opin Anaesthesiol, 2019; 32:129–135.

20.

Moeckli

, Cram

, Cunningham

, Reisinger

. Staff acceptance of a telemedicine intensive care unit program: A qualitative study. J Crit Care, 2013; 28:890–901.

21.

Lilly

, McLaughlin

, Zhao

, Baker

, Cody

, Irwin

, et al. A multicenter study of ICU telemedicine reengineering of adult critical care. Chest, 2014; 145:500–507.

22.

Caples

SM.

Intensive care unit telemedicine care models. Crit Care Clin, 2019; 35:479–482.