Clinical impacts of an integrated electronic health record-based smoking cessation intervention during hospitalisation

Abstract

Objective

To assess the effects of an electronic health record (EHR) intervention that prompts the clinician to prescribe nicotine replacement therapy (NRT) at hospital admission and discharge in a large integrated health system.

Design

Retrospective cohort study using interrupted time series (ITS) analysis leveraging EHR data generated before and after implementation of the 2015 EHR-based intervention.

Setting

Kaiser Permanente Northern California, a large integrated health system with 4.2 million members.

Participants

Current smokers aged ≥18 hospitalised for any reason.

Exposure

EHR-based clinical decision supports that prompted the clinician to order NRT on hospital admission (implemented February 2015) and discharge (implemented September 2015).

Main outcomes and measures

Primary outcomes included the monthly percentage of admitted smokers with NRT orders during admission and at discharge. A secondary outcome assessed patient quit rates within 30 days of hospital discharge as reported during discharge follow-up outpatient visits.

Results

The percentage of admissions with NRT orders increased from 29.9% in the year preceding the intervention to 78.1% in the year following (41.8% change, 95% CI 38.6% to 44.9%) after implementation of the admission hard-stop intervention compared with the baseline trend (ITS estimate). The percentage of discharges with NRT orders increased acutely at the time of both interventions (admission intervention ITS estimate 15.5%, 95% CI 11% to 20%; discharge intervention ITS estimate 13.4%, 95% CI 9.1% to 17.7%). Following the implementation of the discharge intervention, there was a small increase in patient-reported quit rates (ITS estimate 5.0%, 95% CI 2.2% to 7.8%).

Conclusions

An EHR-based clinical decision-making support embedded into admission and discharge documentation was associated with an increase in NRT prescriptions and improvement in quit rates. Similar systemic EHR interventions can help improve smoking cessation efforts after hospitalisation.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This was a multicentre study with a large cohort of patient hospital encounters (41 805) in an integrated health system.

• This study describes standardised, easy-to-implement electronic health record interventions to improve smoking cessation after hospitalisation.

• Even though the use of interrupted time series analyses forms a natural experiment, a limitation could be missed confounders that could introduce bias.

• Another limitation could be that patient-reported quit data may overestimate quit rates.

Introduction

Tobacco use continues to be one of the primary causes of morbidity and mortality worldwide.^{1 2} Cigarette smoking prevalence has been decreasing in the USA, but 50.6 million Americans continue to use tobacco products as of 2019.³ Recent estimates show a 10% reduction in the USA smoking rate would result in $63 billion in savings in healthcare costs 1 year later.⁴ Hospitalisation is a critical touchpoint for smoking cessation. In 2012, the Joint Commission implemented the Tobacco Cessation Performance Measure-Set,⁵ renewing interest in improving the documentation of smoking status and treatments offered throughout health systems nationwide.

Pharmacotherapy, including nicotine replacement therapy (NRT), is effective in increasing smoking cessation.^{6 7} When compared with other pharmaceutical modalities such as bupropion or varenicline, combination NRT has been shown to be equally effective.⁸ Without a concerted effort to increase prescriptions for smoking cessation medications, prescription rates for these medications during hospitalisation and after hospital discharges have been low.⁹ There is limited evidence for methods to increase prescriptions for smoking cessation medications. More recent studies include randomised trials of postdischarge decision support tools, embedded in the electronic health record (EHR), with some improvement in prescriptions but no effect on quit rates themselves.^{4 10 11} In addition, a qualitative analysis of a suite of decision-making support tools aimed at hospital-based physicians to improve smoking cessation treatment has shown that time-consuming methods were less likely to be widely used.¹²

Clinical decision support embedded into EHRs has been recently used to standardise clinical workflows. For instance, a hard-stop intervention that prompts the clinician to take action before allowing them to move forward with their EHR work has been shown to affect clinical behaviour.¹³ A systemic approach to improving support for smoking cessation may be particularly helpful as it is a chronic and relapsing condition.¹⁴ In order to improve the standardisation of smoking cessation treatment during hospitalisations, Kaiser Permanente Northern California (KPNC) implemented a systemwide EHR hard-stop in 2015 to prompt clinicians to order NRT for current smokers during hospitalisation and at discharge. The primary aim of this study was to evaluate the association of this hard-stop EHR intervention with changes in NRT orders during hospitalisation and at discharge during this natural experiment. We also evaluated whether the EHR intervention was associated with patient-reported quit rates 30 days after discharge.

Methods

Setting

This study was conducted within KPNC, a non-profit, large, vertically integrated health system with a fully integrated EHR providing comprehensive healthcare services to 4.5 million members in Northern California across 21 hospitals.¹⁵ KPNC serves a racially, ethnically and socioeconomically diverse population representing 51% of the health insurance market share in California.¹⁶ Due to the vertically integrated nature of the health system, KPNC can provide full-service outpatient and inpatient tobacco control measures and follow-up. In 2015, KPNC implemented systemwide hard-stop EHR interventions in the form of order-sets recommending the orders of NRT on hospital admission in February 2015 and on hospital discharge in September 2015. Clinicians could choose to bypass the order-set completely by indicating that they will return to it at a later time. If they did not want to prescribe NRT, they would have to choose one of the prespecified choices for not ordering NRT as described in the Results section. There was no explicit documentation about the patient’s willingness to quit other than the reason for not ordering NRT. The admission order-set and the discharge order-set were implemented at two separate points in time as noted above and are thus treated as two separate interventions. NRT prescriptions in KPNC during this period did not require copays by patients. Smoking cessation supports during admission and at discharge were separately used by the Joint Commission for hospital accreditation purposes.

Study design

We used an interrupted time series (ITS) design, one of the strongest quasiexperimental study designs, to quantify changes in levels and trends in NRT orders and quit rates during the 24 months before and after implementation of the intervention at KPNC.¹⁷ Inclusion criteria were (1) KPNC adult members aged 18 years and older who were hospitalised for any reason, (2) Kaiser Permanente health plan membership at the time of hospital admission and during the following 30 days and (3) current smokers per self-reported data in the EHR within 6 months before or during hospital admission. Patients with prior NRT orders before their hospitalisation were included. Patients enrolled in hospice, who died or who eloped during their admission were excluded. All hospitalisations during the study period were included. This study was approved by the KPNC Institutional Review Board.

Patient and public involvement

There was no patient and public involvement in the design of this study.

Outcomes and exposures

The primary outcomes were the percentage of patients receiving NRT orders during admission and at discharge. NRT orders included choices between FDA-approved nicotine patches and gum based on the number of packs per day per the patient’s reported smoking history. Discharge NRT orders included new prescriptions as well as continued preadmission prescriptions. The secondary outcome was the percentage of patients reporting new quits within 30 days following discharge (where the discharge order-set implementation was used as the change point for ITS and pre-post modelling). In KPNC, postdischarge outpatient appointments are automatically scheduled. Medical assistants consistently screened for smoking status during the visit and any change in the reported smoking status in the EHR from ‘current smoker’ to ‘former smoker’ was noted as a new quit. Due to the integration at KPNC, all new quits were recorded directly into the EHR. During the postintervention period, in cases where NRT was not ordered, the documented reason why NRT was not ordered was noted. Bupropion and varenicline prescription data were assessed during the study period as a means for controlling background trends for changes in systemic smoking cessation interventions in the health system as they are used in smoking cessation regimens at KPNC but are not in this specific EHR intervention. Data on patient age, race/ethnicity, sex, census-based neighbourhood household median income and Medicaid enrolment were extracted from the EHR.

Statistical analysis

An ITS design estimates the change from preintervention to postintervention in an outcome and tests whether this change is more than would be expected if the intervention never occurred. For each month during the study period, a percentage was derived by counting all outcomes (the numerator) among all hospitalisations (the denominator). The months were divided into preimplementation and postimplementation periods following a 2-year period before and after each order-set intervention. For discharge NRT orders and quit attempts, the 2 years before the discharge intervention (1/9/2013 to 31/8/2015) served as the preimplementation period. The 2 years afterwards served as the postimplementation period (1/9/2015 to 30/9/2017). For admission NRT, the preimplementation period covered 17 months before the admit hard-stop implementation (1/9/2013 to 31/1/2015) and the postimplementation period covered 31 months after implementation (1/2/2015 to 30/9/2017). The months during and after each intervention (admit: 1/2/2015 to 31/1/2015; discharge: 1/9/2015 to 30/10/2015) were excluded from the analysis to allow lead-in time for the implementation of interventions. We analysed data in SAS, V.9.4 (SAS Institute). Each of the ITS models controlled for both intervention periods as well as for the background trend in each outcome (ie, change in the outcome expected without the interventions). To further account for the influence of concurrent trends in smoking cessation prescribing systemwide, we included a model term for monthly orders of smoking cessation medications not part of the hard-stop order-set (varenicline, bupropion). The simplest models assessed the immediate preimpact to postimpact of the interventions. Additional models checked for a change in slope, or trend, by including an interaction term between the primary intervention and time. More details are included in online supplemental materials. For the quit attempt models, patients were excluded from the formal model if they did not have smoking status recorded in the 30 days following inpatient discharge. However, sensitivity analyses were conducted to assess the effect of this exclusion.

Bivariate statistics were used to determine whether there were any population-level changes in the cohort before or after intervention. χ² tests were used for categorical variables (gender, race/ethnicity, Medicaid enrolment, comorbidities, non-NRT smoking cessation medications), and either Student’s t-tests or Kruskal-Wallis tests for continuous variables (age, neighbourhood median income). A pre-post bivariate method with χ² tests was used to assess absolute changes in the primary and secondary outcomes over the year prior to the intervention (1/1/2014 to 31/12/2014) as compared with the year following the intervention (1/1/2016 to 31/12/2016). Poisson regression was performed to account for changes in the distribution of patient characteristics that could contribute to the change in outcome trends around intervention implementation. Further descriptive analysis included reporting frequencies for the reason chosen by the clinician for not ordering smoking cessation medication if the order was omitted.

Results

Study population characteristics

A total of 41 805 patient encounters were identified with 20 847 encounters in the preintervention period and 20 958 encounters in the postintervention period with 29 245 unique patients represented. The median age of the population was 58 years. A majority of the population was non-Hispanic white (60.9%), and the remainder of the population was African American (14.6%), Hispanic (11.5%) and Asian/Pacific Islander (7.5%) (table 1). A total of 4565 (10.9%) of the population had a preceding NRT prescription in the 60 days prior to admission with a significant increase in this measure in the postintervention period from 8.4% to 13.5%.

Table 1.

Patient sociodemographic characteristics and smoking cessation pharmacotherapy prescriptions in the hospitalised patient population studied before and after the hard-stop interventions

Characteristic	Total N (%)	Pre N (%)	Post N (%)	P value*
Race				0.96
Non-Hispanic white	25 460 (60.9)	12 735 (61.1)	12 725 (60.7)
Asian	3130 (7.5)	1552 (7.4)	1578 (7.5)
African American	6114 (14.6)	3034 (14.6)	3080 (14.7)
Hispanic	4793 (11.5)	2378 (11.4)	2415 (11.5)
Other/missing/multi?	2308 (5.5)	1148 (5.5)	1160 (5.5)
Nicotine replacement therapy ordered in last 60 days	4565 (10.9)	1745 (8.4)	2820 (13.5)	<0.001
Woman	19 514 (46.7)	9787 (47.0)	9727 (46.4)	0.27
Age—median (IQR)	58 (47–68)	58 (47–68)	59 (47–68)	<0.001
Neighbourhood median income (100K)—median (IQR)	71.7 (50.9–98.0)	73.3 (51.8–99.3)	70.1 (50.0–95.9)	<0.001
Medicaid enrolment	4391 (10.5)	1911 (9.2)	2480 (11.9)	<0.001
Bupropion prescriptions at discharge	763 (1.8)	277 (1.3)	486 (2.3)	<0.001
Varenicline prescriptions at discharge	79 (0.2)	25 (0.1)	54 (0.3)	0.001

*P values determined using χ² tests for categorical variables and Kruskal-Wallis tests for continuous variables.

NRT orders and prescriptions

An acute increase in admission NRT orders was noted between January and March 2015 after the admission intervention (figure 1A). The ITS model estimated an immediate absolute increase of 41.8% in admit NRT orders (95% CI 38.6% to 44.9%) from 29.9% in the year preceding the intervention to 78.1% in the year following the intervention that was not accounted for by any background upwards trend. Prior to the implementation of the admit intervention, admit NRT orders were increasing by 0.5% per month, but this upwards slope flattened postintervention. Over the year following the intervention (2016) as compared with the year before the intervention (2014), there was an overall 48.1% increase in NRT prescriptions on admission from 3110 (29.9%) to 7886 (78.0%), (p<0.001), with a 2.57 times relative risk increase in likelihood of having NRT admission prescriptions after the intervention (95% CI 2.49 to 2.66) (table 2). Patients aged less than 65 years, of white race and with a previous smoking cessation prescription in the previous 60 days had a significantly increased probability of having an NRT prescription in their medication record over the cumulative study period (table 2).

Figure 1.

Changes in percentage of admissions with (A) nicotine replacement therapy (NRT) ordered during admission, (B) NRT ordered at discharge and (C) patient-reported quits within 30 days following discharge, including interrupted time series (ITS) linear model trend line notes. Dotted line indicates primary intervention date for each outcome. The ITS model used to generate the trend line included a continuous term for month, which controlled for any background trend not attributed to the interventions. To further account for the influence of concurrent trends in smoking cessation prescribing systemwide, we included a term for monthly orders of smoking cessation medications not included in the hard-stop order-set (varenicline, bupropion).

Table 2.

Association of patient characteristics and changes in risk of nicotine replacement therapy prescription and posthospitalisation quits after discharge hard-stop intervention using multivariable Poisson regression

Characteristic	Admit NRT	Discharge NRT	Patient-reported quits
	RR (95% CI)	RR (95% CI)	RR (95% CI)
Postintervention (ref pre)	2.57 (2.49 to 2.66)	2.8 (2.68 to 2.93)	1.26 (1.19 to 1.33)
Age <65 (ref 65+)	1.07 (1.04 to 1.10)	1.01 (0.97 to 1.05)	0.90 (0.84 to 0.95)
Race (ref white)
Asian	0.89 (0.85 to 0.94)	0.96 (0.90 to 1.03)	1.42 (1.31 to 1.55)
African American	0.92 (0.89 to 0.96)	1.04 (0.99 to 1.09)	0.89 (0.82 to 0.98)
Hispanic	0.90 (0.86 to 0.93)	0.88 (0.83 to 0.93)	1.27 (1.18 to 1.38)
Other	0.94 (0.89 to 0.99)	1.00 (0.93 to 1.07)	0.97 (0.85 to 1.10)
Male gender (ref female)	1.02 (0.99 to 1.04)	0.97 (0.94 to 1.01)	1.05 (1.00 to 1.11)
Medicaid insurance	0.99 (0.96 to 1.03)	1.00 (0.95 to 1.05)	0.80 (0.72 to 0.89)
Neighbourhood median income (ref 120 000+)
<40 000	1.03 (0.99 to 1.07)	1.05 (1.00 to 1.11)	1.00 (0.92 to 1.08)
40 000-<120 000	1.00 (0.95 to 1.05)	1.06 (0.99 to 1.14)	1.00 (0.89 to 1.11)
Smoking cessation Rx in 60 days prior to admit	1.24 (1.21 to 1.28)	ND	1.04 (0.95 to 1.13)

Note: Multivariable models adjust for each characteristic shown in the table.

ND, no data available since we studied new NRT prescriptions after discharge rather than continued prescriptions for patients who have NRT prescriptions prior to their hospitalisation; NRT, nicotine replacement therapy.

Prescriptions of NRT at discharge had a more gradual increase over the year of the order-set implementations. There was an acute increase in NRT prescriptions at discharge after the implementation of the admission order-set (ITS estimate 15.5%, 95% CI 11.0% to 20.0%) with another increase associated with the implementation of the discharge order-set (ITS estimate 13.4%, 95% CI 9.1% to 17.7%). The final discharge NRT model accounted for first-order autocorrelation of data points and a change in trend/slope prehard-stop to posthard-stop implementation (0.6% increase per month preimplementation vs 0.2% increase per month postimplementation) (figure 1B). There was a 39.5% absolute increase in NRT orders filled at discharge between 2014 and 2016 from 1345 (21.9%) in 2014 to 4621 (61.4%) in 2016 (p<0.001), with a significantly increased likelihood of NRT prescriptions filled at discharge following the intervention (RR: 2.8, 95% CI 2.68 to 2.93). Patients reporting Hispanic ethnicity had a lower likelihood of having NRT discharge prescriptions (table 2).

New quit reports after discharge

Overall, 82% of the population had a new smoking status reported in the month following their hospitalisation, with a majority of these reports occurring in outpatient visits following discharge. Patients who did not have a follow-up smoking status change were younger, were more likely to be women, African American, have Medicaid insurance, and have received NRT on discharge. Following the discharge intervention (but not the admit intervention), there was a small but significant spike in new self-reported quits (ITS estimate 5.0%, 95% CI 2.2% to 7.8%) (figure 1C). To account for any selection bias caused by the exclusion of patients without any smoking status noted in the follow-up visit, we conducted a sensitivity analysis that included all patients in the quit attempt analysis (with smoking status, without smoking status or without follow-up) with similar results, showing a significant increase in quit attempts in the postimplementation period (ITS estimate 4.0%, 95% CI 1.6% to 6.4%). The ITS model suggested no background trend in new quits in either the preintervention or postintervention periods, but there were significant autocorrelation terms at the first and seventh order, suggesting a potential seasonal cycle for changes in new quits over the year. In the pre-post analysis, the rates of patient-reported new quits increased from 2077 (24.5%) in 2014 to 2564 (30.7%) in 2016 (p<0.001) (figure 2). An increased likelihood of new quits following the intervention persisted after controlling for covariates (RR, 1.26; 95% CI 1.19 to 1.33). Patients who reported quitting had a significantly higher percentage of discharge NRT than patients who reported continued smoking (54.0% vs 43.0%, data not shown). Patients with Asian or Hispanic race/ethnicity as compared with white race were more likely to report new quits (table 2). Patients with age less than 65 years, African American race and Medicaid enrolment were significantly less likely to report quitting in the 30 days following their hospitalisation (table 2).

Figure 2.

Percentage of patients with nicotine replacement therapy orders and quit attempts over the year before versus after the intervention (2014 vs 2016) note. All χ² p<0.001.

Reasons for not ordering NRT at discharge

If NRT was not ordered at the point of discharge after the intervention was implemented, clinicians were prompted to record the reason for not ordering NRT in order to bypass the hard-stop. However, for 3050 (16.0%) of the discharge instances in the postperiod, the clinicians were able to discharge the patients without indicating a reason for not ordering NRT by choosing the option to return to the order-set at a later time without doing so (ie, bypassing the hard-stop) (figure 3). Of the 4761 instances when NRT was not ordered and the clinician reported a reason, the majority of encounters noted patient declining as the reason for not ordering NRT (3552 (18% of encounters in postperiod)). The other options that could be chosen included already ordered outside order-set (this may have included previously filled medications, 220 (1%)), pregnancy or breastfeeding (153 (1%)), cardiac risk (126 (1%)) or ‘other’ (710 (4%)) (figure 3).

Figure 3.

Percentage of hospitalisations with nicotine replacement therapy ordered at discharge and clinician-reported reason for bypass of order-set during the full postintervention period (September 2015 to September 2017) note. Order-set over-ride is defined as the case when clinicians bypass the use of the order-set by choosing to return to the order-set at a later time during the discharge process but failing to do so.

Discussion

A systemwide EHR hard-stop intervention that prompts clinicians to order NRT during hospital admission and discharge was successful in changing clinician behaviour and subsequently led to significant increases in NRT prescription orders and patient-reported new quit rates. This study is among the first to show a health systemwide EHR intervention can improve smoking cessation therapy prescription rates associated with hospitalisations. Results are consistent with findings from a pharmacist-led opt-out cessation treatment integrated into existing inpatient medical reconciliation workflows.¹⁸ Prior evidence has shown that NRT is underprescribed during hospitalisations regardless of the reason for admission.¹⁹ A 2014 Cochrane review of EHR-embedded smoking cessation tools found no studies that showed an effect on prescriptions for smoking cessation medications.¹¹ Interestingly, the largest increase in NRT prescriptions on discharge was associated with the implementation of the admission order-set, suggesting that once NRT was ordered as part of the admission orders, clinicians were more likely to continue it upon discharge. The implementation of the discharge order-set in addition had an additional increase in discharge NRT orders as shown in figure 1 through the interrupted time series design. This type of EHR mechanism can ensure equity in types of care delivery and simplify clinical practices by standardising work in a population health setting.²⁰ Smoking cessation interventions during hospitalisations have been shown to be cost-effective,²¹ and this type of simple intervention built into an existing EHR can lead to even further cost-savings.

The small yet significant improvement in new quits over the month following hospitalisations is very encouraging. Due to the integrated study setting, 82% of hospitalisations had a follow-up outpatient visit in the 30 days following discharge where smoking status was recorded. Quits increased acutely after implementation of the discharge order-set only, which suggests the cumulative increase in NRT prescriptions at discharge from both order-sets could be especially helpful in promoting quitting after leaving the hospital. This conclusion is also supported by the finding that patients who reported new quits also had significantly higher rates of discharge NRT and not admission NRT. This EHR intervention led to an increased rate of NRT prescriptions at discharge. However, clinicians were able to work around the intervention as observed rates of NRT prescriptions were not universal despite the goal of this type of EHR hard-stop to standardise clinician behaviour and reduce bias. The ability of the clinician to forego ordering the NRT was crucial to ensure that clinical discretion could be used to ensure that the patient received appropriate care. However, clinicians may be choosing to bypass the ordering of NRT volitionally as part of their clinical judgement or in error by choosing options that noted that the NRT was ordered separately or patient refused when that may not have been the case. Additional studies that examine clinician-level factors that may impact NRT orders in response to an EHR-based intervention are needed. A recent study in China suggested a decrease in further admissions for cardiovascular events after the implementation of a multipronged smoking cessation approach using an interrupted time series analysis.²² However, this is the first study to demonstrate the effectiveness of a simple EHR intervention on not only medication prescription but also new quits. Certain patient characteristics like age were associated with NRT prescriptions.

Interestingly, compared with non-Hispanic white patients, those of every other race and ethnicity noted were less likely to have NRT orders on admission, but only Hispanic patients were less likely to have NRT orders on discharge. It is not clear if this is due to patient or provider preference and further study into this association with race is needed. Further, additional research is needed to investigate other factors associated with NRT prescribing (eg, clinician specialty or reason for patient admission).

Limitations

The limitations of this study include the observational nature of the ITS design as there could be confounders changing at the same time as the interventions that we have not controlled for. The quit rate studied here is patient reported and thus could be falsely high due to reporting bias.²³ However, since follow-up was not universal, we could have also undercounted the overall quits. Patients could start smoking again after the 30-day period, and thus the quit rate reported is not evidence of permanent smoking cessation. Data from patients excluded including those who died during their admission could not be assessed and could have led to an inflation of quit rates reported; however, they would not have been eligible for the intervention studied. We were not able to assess whether patients actually used the NRT that was ordered (eg, patients could have declined NRT at the bedside). Further, the lack of data on the number of doses given or the duration of treatment is also a limitation. Additional studies are needed to see whether the benefits of this EHR intervention are sustained over long periods of time or translatable to other settings.

Strengths

The strengths of this study include the large populationwide assessment with an interrupted time series design of a simple intervention to improve smoking cessation support prescriptions. In the absence of a natural control group (ie, patients excluded from the intervention), the counterfactual (what would have happened in the absence of the policy) is estimated using stable baseline trends. The inclusion of drugs not included in the intervention as a means of controlling for prescribing trends adds to the strength of this type of assessment of a natural experiment. This EHR intervention may not be as easily replicable in settings without an integrated, robust EHR.

Conclusions

In the USA, less than a third of smokers trying to quit use evidence-based smoking cessation strategies such as NRT.²⁴ Tobacco cessation treatment at the time of hospital discharge is the standard of care.⁸ A simple EHR hard-stop intervention can greatly improve smoking cessation therapy prescriptions during hospitalisations and may be associated with a small yet significant improvement in smoking cessation in this diverse population studied. In this cohort, the admission EHR intervention was associated with an increase in discharge NRT prescriptions; however, the medications on discharge seemed to be more strongly associated with new smoking quits rather than the NRT orders during admission themselves. Therefore, prioritising EHR interventions to ensure discharge NRT should be a goal for other healthcare systems implementing similar interventions. A successful smoking cessation intervention during hospitalisation may be especially apt as it targets a vulnerable population at high risk of significant comorbidities.

Data availability statement

No data are available.

Ethics statements

Patient consent for publication

Not applicable.

Ethics approval

This study involves human participants but this study was reviewed as exempt by the Kaiser Permanente—Northern California Region IRB, Oakland, CA - 1502251-3. This study involved secondary research of identifiable private information for which consent is not required because the research involves only information collection and the analysis of this identifiable health information is regulated by theHealth Insurance Portability and Accountability Act (HIPAA) in the United States of America. There was no explicit consent process for this study as data were used in secure processes as regulated through HIPAA.