Automated Screening of Hospitalized Children for Influenza Vaccination

Abstract

Background

This study was designed to determine whether an automated hospital-based influenza vaccination screening program leveraging the electronic medical record (EMR) increases vaccination rates.

Methods

We performed a retrospective cohort study of all children ≥6 months old admitted to medical, surgical, rehabilitation, or psychiatry services during influenza seasons between 2003 and 2012 at a tertiary care pediatric hospital. We compared influenza vaccination rates before (preintervention phase) and after (intervention phase) the introduction of an automated EMR intervention that utilized a nursing-based electronic screening tool to determine eligibility for influenza vaccine and facilitated vaccine ordering without requiring involvement of a physician or other provider.

Results

Overall, 42 716 (72.8%) of the 58,648 subjects admitted during the study period met inclusion criteria. The intervention phase included 20,651 admissions, of which 11 194 (54.2%) were screened. Screening increased significantly over time in the intervention phase (19.8%–77.1%; P < .001). In-hospital influenza vaccination rates increased from a mean of 2.1% (n = 472) of all subjects preintervention phase to 8.0% (n = 1645) in the intervention phase (odds ratio = 6.8; 95% confidence interval, 6.14–7.47). Of the 11 194 screened subjects, 5505 (49.2%) were found to have already been vaccinated at the time of screening. The screening process identified 478 (4.3%) subjects who were unable to receive vaccine for medical reasons, and an additional 2865 (25.6%) whose caregiver refused the vaccine.

Conclusions

An automated, hospital-based influenza vaccination program integrated into the EMR can increase vaccinations of hospitalized patients and provide insight into the vaccination history and declination reasons for children not receiving the vaccine.

The Centers for Disease Control and Prevention (CDC) currently recommends that all children ≥6 months of age receive the influenza vaccine annually [1–3]. The CDC recommendations for influenza vaccination have expanded over the last decade: in the 2003–2004 season, children aged 6–23 months were encouraged to the receive the vaccine; in the following year, it became a formal recommendation [1, 4]; in the 2006–2007 season, the recommendations were expanded to include children aged 24–59 months; and in the 2008–2009 season, they were expanded further to include all children ≥6 months [5].

Influenza infects people of all age groups, although the highest morbidity and mortality is seen in adults ≥65 years, children ≤2 years of age, and those with certain chronic medical conditions [6–8]. Only an estimated 43% of children between 6 months and 17 years received the influenza vaccine in the 2009–2010 season [9]. Although high-risk children have slightly higher vaccination rates, a substantial number still go unvaccinated annually [10]. Furthermore, a significant number of children hospitalized for influenza-related illnesses have been hospitalized previously in the same season, providing an opportunity for vaccination [11].

In order to vaccinate all children as recommended, new vaccination strategies incorporating venues outside the medical home are needed. Standing-order programs have successfully increased influenza vaccination rates for adults in the hospital setting and have also been advocated for children [12, 13]. In addition, clinical decision support (CDS) systems have been shown to increase vaccinations in a variety of settings with varying success [14–16]. However, most CDS tools have focused on the ordering provider workflow, although some preliminary work investigated the potential for influencing nursing-based practices as well [17, 18].

Our objective was to examine whether an automated hospital-based influenza vaccination program leveraging the electronic medical record (EMR) directed towards nursing increased influenza vaccination rates for all qualified patients. We hypothesized that automated screening and vaccination would improve influenza vaccination rates among hospitalized children.

METHODS

Overview

We performed a retrospective cohort study of all admissions within the typical influenza vaccination and illness season (October 1 through April 30) during the 2003–2004 to 2011–2012 seasons at Seattle Children's Hospital (SCH). The SCH EMR, Cerner's Powerchart, provides tools that allow an organization to create custom functionality. We developed and implemented an electronic process in the 2008–2009 season that automatically (1) determined potential eligibility for influenza screening based on age and admitting service, (2) provided a mechanism within the EMR to complete the screening process, and (3) ordered an appropriate vaccine formulation, if indicated. The study protocol was reviewed and approved by the SCH Institutional Review Board.

Population

All children ≥6 months of age hospitalized on the medical, surgical, psychiatric, or rehabilitation units were targeted for the screening tool and were included. Children ≥6 months of age transferred from the intensive care unit to one of the targeted units were also included. Children <6 months of age or those admitted to either the intensive care or oncology services were excluded (the oncology service tracked and vaccinated patients internally to control vaccination timing and optimize immune response given the complexities of their patients' immunosuppressive therapies), as were those vaccinated at SCH within the past 30 days, because a second influenza vaccine dose (if indicated) should be separated from the first dose by at least 30 days. We stratified patients into high risk (such as those with cardiopulmonary, autoimmune, or renal disease) or low risk for developing secondary complications from influenza using International Classification of Diseases, 9th revision (ICD-9-CM) discharge diagnosis codes, as has been done previously [19].

Intervention

We divided the study period into preintervention (2003–2004 through 2007–2008 seasons) and intervention phases (2008–2009 through 2011–2012 seasons). Once a child was flagged for screening at the time of admission or transfer, bedside nurses could choose an appropriate time to complete the screening form, which led them through a standardized sequential process. Children with a fever (≥38.5°C) were deferred and automatically reflagged 12 hours later. The form was updated annually to reflect current CDC recommendations. During the intervention phase, the hospital promoted the seasonal influenza vaccination campaign and reminded nurses of the automated screening tool annually. The form (Appendix 1) screened individual patients through the assessment of both parental or guardian consent as well as medical eligibility to receive vaccine. Once a patient was found eligible, the form guided the child's assigned nurse to select live or inactivated influenza vaccine. After the nurse selected the desired vaccine, the system automatically placed the appropriate order without requiring direct intervention from a physician or other provider. Beginning in the 2008–2009 season, all elements of the electronic screening form were captured, including reasons for vaccine ineligibility (ie, allergy to vaccine, etc) and parental or guardian declination. All vaccine orders and administration records were captured within the EMR due to mandatory computer provider order entry at SCH during the entire study. Furthermore, beginning in 2009, SCH reported all vaccine administrations to the Washington State vaccine registry, including historical records dating back to 2005.

Outcomes

The two primary outcomes considered were screening status (screened vs unscreened) and vaccination status (vaccinated vs unvaccinated). Although the process was automated, some subjects did not have screening completed, such as those admitted with a fever and discharged shortly thereafter, or those whose screening tasks were not completed by their bedside nurses. Screened subjects had the potential for one of the following vaccination outcomes: medically ineligible, caregiver refusal, previously vaccinated during the current season (by parental or caregiver report), vaccine ordered but not administered, or vaccine ordered and administered. Unscreened subjects had 2 possible outcomes: vaccinated at SCH or unvaccinated at SCH. Vaccine administration was confirmed using the computerized medication administration record, which was utilized during the entire study period.

Analyses

We summarized screening and vaccination rates using proportions. The χ² test was used to compare rates between preintervention and intervention periods, and across groups. The χ² test for trend was used to compare rates across multiple seasons. Multivariate logistic regression was used to assess the associations between screening and vaccination, while adjusting for covariates. Covariates associated with screening or vaccination, respectively, with P < .2 in univariate analysis were included in the initial multivariate model, and the most parsimonious model was selected using a stepwise backward selection process. Data were analyzed using Stata version 12.0 (College Station, TX).

RESULTS

Screenings

Overall, 58 648 patients were admitted during the 9 influenza seasons; 42 716 subjects (72.8%) met inclusion criteria (≥6 months of age and admitted to the medical, surgical, psychiatric, or rehabilitation units). After program implementation in the 2008–2009 season, 27 803 patients were admitted, of whom 20 651 (74.3%) met inclusion criteria (Table 1). During the intervention phase, 11 194 (54.2%) of the 20 651 subjects were screened, and the screening rate increased during the intervention phase from 19.8% to a high of 81.1% (P < .001; Figure 1).

Table 1.

Demographic and Clinical Information for Included Subjects

Total	Preintervention Phase 2003–2008 [n (%)] 22 065 (51.7)	Intervention Phase 2008–2012 [n (%)] 20 651 (48.3)	P Value
Age
6 mo to <2 y	4902 (22.2)	4194 (20.31)
2 y to <5 y	4356 (19.7)	3946 (19.1)
5 y to <12 y	6001 (27.2)	5717 (27.7)
> 12	6806 (30.9)	6794 (32.9)	<.001
Male	11 632 (52.7)	10 950 (53.0)	.5
Race
American Indian	310 (1.4)	321 (1.6)
Asian	1185 (5.4)	1163 (5.6)
Black	1430 (6.5)	1286 (6.2)
White	13 897 (63.0)	12 052 (58.36)
Native Hawaiian or Other Pacific Islander	156 (0.71)	212 (1.03)
Other/Missing	5243 (23.8)	5829 (28.2)	<.001
Ethnicity
Hispanic/Latino	2701 (12.24)	3196 (15.5)
Non Hispanic/Latino	14 945 (67.73)	15 807 (76.5)
Other/Missing	4419 (20.0)	1648 (8.0)	<.001
Service
Medical	12 881 (58.4)	11 669 (56.5)
Surgical	8002 (36.3)	7472 (36.2)
Rehabilitation	263 (1.2)	224 (1.1)
Psychiatry	919 (4.2)	1286 (6.2)	<.001
High-Risk Status	4083 (18.5)	3341 (16.2)	<.001

Figure 1.

Percentage of hospitalized patients screened and vaccinated for influenza by season. Abbreviation: SCH, Seattle Children's Hospital.

Subjects on Medical, Rehabilitation, and Psychiatric teams had an increased chance of being screened compared with those on Surgical teams (54.6%, 65.6%, 72.2%, and 50.1%, respectively). Age, sex, and race were not associated with the likelihood of being screened. Adjusting for year and admitting service, patients at high risk for influenza complications were less likely to be screened (1776 of 3341, 53.2%) than those classified as low risk (9417 of 17 310, 54.4%; odds ratio [OR] = 0.91; 95% confidence interval [CI], .83–.99). Screening was more likely to occur in those whose length of stay was longer than the 2-day, overall median length of stay than in those whose length of stay was shorter than the median (60.6% vs 49.7%; P < .001).

Vaccinations

Of all subjects, 2117 (5%) received a vaccine before discharge (96.3% received trivalent-inactivated vaccine, whereas 3.7% received the intranasal, live-attenuated vaccine). In-hospital influenza vaccination rates increased from a mean of 2.1% (n = 472) of all subjects preintervention phase to 8.0% (n = 1645) in the intervention phase (P < .001; Figure 1). Of the 2396 patients in the intervention phase that were screened and had no contraindications or refusals, 2153 (90%) had vaccine ordered and 1461 (67.9%) received the vaccine. The proportion of subjects with a vaccine order who received a vaccination decreased over time from 89.1% in the 2003–2004 season to 63.6% in the 2011–2012 season (P < .001). An additional 184 (1.9%) of the 9457 unscreened patients received a vaccine that was ordered by a physician or other provider outside of the automated process. Factors associated with in-hospital vaccination in univariate and multivariate models are presented in Table 2; medical service and ethnicity significantly influenced the likelihood of being vaccinated within the hospital. Increased length of stay was associated with increased likelihood of vaccination (OR = 1.01; 95% CI, 1.00–1.01). In the fully adjusted model, screening was associated with a 6-fold increase in odds of vaccination during hospitalization (OR = 6.8; 95% CI, 6.1–7.5). The most parsimonious multivariate adjusted model for the association between screening and in-hospital vaccination did not include length of stay; when length of stay was included in the model, the odds of vaccination with screening was no different (OR = 6.8; 95% CI, 6.1–7.5).

Table 2.

Factors Associated With Influenza Vaccination During Hospitalization Between 2003 and 2012

	Univariate		Multivariate
	OR (95% CI)	P Value	OR (95% CI)	P Value
Automated Screening	7.06 (6.42–7.77)	<.001	6.77 (6.14–7.47)	<.001
Sex
Male	1	-	1	-
Female	0.89 (.82–.98)	.016	0.88 (.80–.96)	.005
Race
American Indian or Alaska Native	1	-	1	-
Asian	0.91 (.60–1.37)	.60	0.93 (.61–1.41)	.719
Black	1.12 (.75–1.66)	.59	1.11 (.73–1.67)	.628
White	0.93 (.66–1.35)	.71	0.95 (.65–1.39)	.792
Native Hawaiian or Other Pacific Islander	1.97 (1.19–3.26)	.008	1.95 (1.15–3.30)	.013
Other	1.22 (.83–1.77)	.31	1.00 (.67–1.50)	.981
Ethnicity
Hispanic/Latino	1	-	1	-
Non-Hispanic/Latino	0.77 (.69–.87)	<.001	0.77 (.66–.89)	.001
Service
Surgical	1	-	1	-
Medical	1.75 (1.58–1.95)	<.001	1.79 (1.60–1.99)	<.001
Rehabilitation	4.65 (3.53–6.11)	<.001	4.27 (3.19–5.72)	<.001
Psychiatry	2.56 (2.14–3.05)	<.001	1.68 (1.39–2.02)	<.001
Age
6 mo to <2 y	1	-	1	-
2 y to <5 y	1.07 (.93–1.24)	.35	1.10 (.94–1.27)	.244
5 y to <12 y	1.21 (1.07–1.39)	.004	1.26 (1.10–1.44)	.001
> 12	1.40 (1.23–1.58)	<.001	1.44 (1.26–1.65)	<.001
Length of Staya	1.01 (1.00–1.01)	<.001	-	-
High-Risk Status	0.78 (.69–.88)	<.001	0.77 (.67–.87)	<.001

Abbreviations: CI, confidence interval; OR, odds ratio.

^aLength of stay was not included in the most parsimonious multivariate model. Including length of stay in the multivariate model did not alter the relationship between screening and vaccination and did not improve the fit of the model.

The screening process identified 5505 (49.2%) subjects who had already received the influenza vaccine previously in the same season. The rate of those previously vaccinated increased over the intervention phase from 44.5% to 54.5% (P < .001) (Figure 1).

The majority of subjects (35 292, 82.6%) were identified as low risk. Of the 17 982 low-risk subjects during the preintervention phase, 400 (2.2%) were vaccinated during hospitalization, compared to 1416 (8.2%) of 17 310 low-risk subjects during the intervention phase (P < .001). The rate of vaccination among unscreened low-risk subjects in the intervention phase was 135 of 7758 (1.7%), lower than the 2.2% vaccinated in the preintervention phase (P = .007). When considering the 7424 total high-risk subjects, 72 (1.8%) of 4083 were vaccinated during hospitalization preintervention and 229 (6.9%) of 3341 were vaccinated postintervention (P < .001) (Figure 2). Postintervention, high-risk subjects were less likely to be vaccinated within the hospital compared to low-risk subjects at 6.9% versus 8.2% (P = .01), but they had a higher likelihood of having been vaccinated prior to hospitalization at 59.8% versus 47.2% (P < .001). Combining those vaccinated prior to hospitalization with those vaccinated during hospitalization, 69.5% of high-risk subjects were vaccinated before discharge compared to 60.6% of those classified as low-risk (P < .001).

Figure 2.

Vaccination status by season and risk level. Abbreviation: SCH, Seattle Children's Hospital.

Likewise, older children and adolescents were more likely to be vaccinated within the hospital compared to children under 2 years of age (Table 2). However, children under 2 years had a higher likelihood of having been vaccinated prior to hospitalization (58.5% vs 46.9%, P < .001) (Figure 3). In addition, there was a significant increase in vaccinations among those under 2 years observed in the intervention phase compared to the preintervention phase (269 of 4194 or 6.4% vs 110 of 4902 or 2.2%; P < .001). This trend was not observed in the unscreened children under 2 years during the intervention phase (52 of 1995 or 2.6% vs 2.2%; P = .34).

Figure 3.

Proportion of screened subjects vaccinated at Seattle Children's Hospital or prior to admission, by age. Abbreviation: SCH, Seattle Children's Hospital.

Declinations

A total of 478 (4.3%) of the 11 194 screened patients were ineligible to receive the vaccine for medical reasons, and an additional 2865 (25.6%) had the vaccine refused by their caregiver (Table 3). Medical contraindications increased from 2.8% to 6.4% over the screening period (P < .001), whereas caregiver refusals declined over time from 30.3% to 23.2% (P < .001).

Table 3.

Reasons for Medical Ineligibility or Caregiver Refusal of Influenza Vaccine

	N (% of screened subjects)
Medical Ineligibility	478 (4.3)
Fever	179 (1.6)
Scheduled Surgery within 48 hours	112 (1.0)
Allergy	95 (0.8)
Primary service denieda	55 (0.5)
Oncology Patient	36 (0.3)
History of Guillain-Barre Syndrome	1 (0.009)
Caregiver refusal	2865 (25.6)
Parent thinks vaccine unnecessary	723 (6.5)
Prefers to receive vaccine elsewhere	490 (4.4)
Parental concerns about vaccine	333 (3.0)
Parents want to wait	327 (2.9)
Parents not interested at this time	198 (1.8)
Other	794 (7.1)

^aThe primary team caring for the patient determined that the patient should not receive the vaccine for unspecified medical concerns; these reasons were not captured on the screening form.

DISCUSSION

Our large, retrospective cohort study of children hospitalized at a pediatric tertiary care center demonstrates that influenza vaccination rates of hospitalized patients increased after the introduction of an automated hospital-based influenza vaccination program. Furthermore, such a process provided insight into both whether children had received the vaccine previously and the reasons why they chose not to receive the vaccine.

Factors that contributed to the success of this program included its being self-contained and nursing-centered. Because the screening process was built into the EMR with automated triggers and alerts, nurses needed to make no decisions to determine screening eligibility. Furthermore, the screening form contained all necessary questions and tasks to be completed in 1 standardized workflow. The form was self-explanatory, and changes in clinical screening questions (as occurred in recent years with the 2009 H1N1 virus, for example) could be made without the need for extensive communication campaigns for the screening staff. Most importantly, ordering a vaccine was facilitated by not requiring action by a physician or other provider.

It is interesting to note that younger patients were less likely to be vaccinated within the hospital, although we found that they were also more likely to have been vaccinated prior to their hospitalization. Current American Academy of Pediatrics and Bright Futures guidelines suggest that children between the ages of 6 months and 3 years see their primary care provider (PCP) 8 times for preventative healthcare [20]. This schedule provides ample opportunity for influenza vaccination, because it is very likely that a proportion of these visits will occur during influenza season. Comparatively, children over the age of 3 usually have annual visits with their PCP, often around their birthday, and not necessarily during influenza season. This requires families to schedule separate appointments for the sole purpose of receiving the vaccine, leading to additional costs and inconveniences.

The screening process provided additional information regarding the vaccination status of our population and insight for those patients choosing not to be vaccinated. Before the intervention phase, we were unable to identify patients who received an influenza vaccine prior to admission or reasons why caregivers declined the vaccine, because this information was only captured with the electronic screening tool. Within our organization, Hispanic patients were more likely to be vaccinated, conflicting with previous studies that have looked at vaccine acceptance in minority populations [21, 22]. Patients on the surgical services were less likely to be screened and less likely to be vaccinated in the hospital. Despite low documented rates of postvaccination fever [23], concern about this possibility and potential confusion with perioperative complications likely contributed to this finding. Gust et al [24] demonstrated that the reasons for vaccine refusals are not homogenous and require individualized response. In this regard, caregivers may have used medical contraindication or other nonspecific reasons as a way of avoiding discussing concerns about the vaccine with care providers. Our automated screening tool exposed differences in vaccination rates between different populations and can therefore be used to develop strategies to address these gaps at our own institution.

Now that US News and World Report and the Centers for Medicare and Medicaid Services include influenza vaccination in their reporting, institutions could implement systems like ours to identify and target both undervaccinated groups as well as undervaccinating providers. This new information will undoubtedly provide opportunities to fine-tune vaccination strategies for organizations as well as expand the opportunities for vaccination. Screening processes are not without their limitations, however. In our program, screening occurred at admission and vaccination was recommended just prior to discharge. Those with shorter lengths of stay were less likely to have screening performed; furthermore, the period of time between vaccine order and delivery may have contributed to the number of patients with vaccine orders but who did not receive a vaccine, in that the care team at discharge may have missed vaccine orders placed a number of days prior.

The strengths of our study include a multiyear, large sample size and EMR-codified data, which allowed us to show the impact of screenings on vaccination as well as differences between medical services. In addition, the EMR allowed us to combine clinical, administrative, and parental reports, so that we could better identify which patients received the influenza vaccine prior to hospitalization.

Our study does have limitations. First, it is possible that clinical screening happened in some subjects without the use of the automated screening tool in the intervention phase. This type of nonstandardized clinical screening relies on clinician memory of both vaccine recommendations and the need for influenza vaccination, which was the process in place prior to the intervention, and remains the current screening practice for clinicians who do not have the benefit of automated screening. The fact that patients in both the low-risk category and those between the ages of 6 months and 23 months had higher vaccination rates if they were screened, even during the intervention phase, suggests that the automated screening itself leads to improvements in vaccination rates over relying on clinical screening.

Second, the screening tool was not integrated with the state vaccination profile, and therefore we relied on caregiver report to provide history regarding previous vaccinations, which limited our ability to identify and track which subjects needed and appropriately received a second dose of vaccine. However, beginning with the 2008 season, all in-hospital vaccinations were reported to the Washington State vaccination registry and indirectly to the patients' PCPs.

Third, we could otherwise misclassify certain groups (such as race, or those with high-risk medical conditions) on the basis of administrative data, but this would again bias any differences among groups to the null, and we have used previously described definitions for high-risk patients [11]. Although our adjustment for illness severity may have been incomplete, adjusting for length of stay did not alter the relationship between screening and vaccination, suggesting that there was minimal residual confounding of this relationship by illness severity. Finally, secular trends, such as changes over time in vaccination recommendations by the Advisory Committee on Immunization Practices (ACIP) as well as public fear over novel influenza strains such as the 2009 H1N1 virus, may have contributed to our findings. Although the ACIP recommendations remained constant, concerns about 2009 H1N1 may have contributed to the increase in vaccinations seen during the intervention phase. Furthermore, unscreened subjects in the intervention period had a lower rate of vaccination than subjects in the preintervention period, suggesting that secular trends in vaccination did not play a large role in the increase in vaccinations over the study period, and that the increase was more likely due to our automated process.

Development of EMR-based tools, such as the one described herein, are guided by many stakeholders with different priorities and multiple constraints, and therefore developing the perfect tool is not always possible [25]. The biggest constraint often is the electronic environment in which the tool is built and used. Different systems have varying capabilities—for example, the ability to autocomplete forms to include allergies, age, or recent vaccine administration—and must be customized to and optimized for the institution in which they are used.

Future study will be important to identify tools that can further increase the rate of screening and vaccinations. Having tools that integrate with state vaccination registries would boost their utility and streamline the process. This model could also be implemented for other needs where large patient populations need to undergo screening as well as generalized health maintenance. For example, having a system integrated with vaccine registries would allow institutions to screen all patients for vaccine deficiencies automatically at the time of admission and potentially provide catch-up vaccinations for these children.

CONCLUSIONS

Implementation of an easy-to-use and integrated automated screening tool can increase influenza vaccination rates, identify those with prior vaccination, and help providers understand reasons for vaccination declination among hospitalized children.

Appendix 1

Electronic medical record (EMR)-based screening tool from 2011 to 2012 season. Tool was designed to allow screening and ordering to occur within a single process. The form leads nurses through a standard workflow for screening as well as provides additional reference information.