Factors motivating and affecting health information exchange usage
- 1Department of Health Policy & Management, School of Rural Public Health, Texas A&M Health Science Center, College Station, Texas, USA
- 2Department of Epidemiology & Biostatistics, School of Rural Public Health, Texas A&M Health Science Center, College Station, Texas, USA
- 3Department of Information & Operations, Mays Business School, Texas A&M University, College Station, Texas, USA
- Correspondence to Dr Joshua R Vest, Jiann-Ping Hsu College of Public Health, Georgia Southern University, Hendericks Hall, PO Box 8015, Statesboro, GA 30460, USA;
- Received 3 April 2010
- Accepted 23 November 2010
- Published Online First 24 January 2011
Objective Health information exchange (HIE) is the process of electronically sharing patient-level information between providers. However, where implemented, reports indicate HIE system usage is low. The aim of this study was to determine the factors associated with different types of HIE usage.
Design Cross-sectional analysis of clinical data from emergency room encounters included in an operational HIE effort linked to system user logs using crossed random-intercept logistic regression.
Measurements Independent variables included factors indicative of information needs. System usage was measured as none, basic usage, or a novel pattern of usage.
Results The system was accessed for 2.3% of all encounters (6142 out of 271 305). Novel usage patterns were more likely for more complex patients. The odds of HIE usage were lower in the face of time constraints. In contrast to expectations, system usage was lower when the patient was unfamiliar to the facility.
Limitations Because of differences between HIE efforts and the fact that not all types of HIE usage (ie, public health) could be included in the analysis, results are limited in terms of generalizablity.
Conclusions This study of actual HIE system usage identifies patients and circumstances in which HIE is more likely to be used and factors that are likely to discourage usage. The paper explores the implications of the findings for system redesign, information integration across exchange partners, and for meaningful usage criteria emerging from provisions of the Health Information Technology for Economic & Clinical Health Act.
Health information exchange (HIE) is the process of electronically transmitting patient-level information between healthcare organizations, and promises to be a solution to the threats posed by the American healthcare system's fragmented approach to health information.1 2 Through HIE, previously inaccessible data become available, resulting in more complete clinical information and potential improvement along nearly all of the Institute of Medicine's quality dimensions3–8 including estimates of billions in cost savings.9 10 Despite the promised benefits, HIE systems historically are utilized for few patients11 12 and few encounters,13 14 and often by a minority of clinicians.15 Furthermore, even though the Centers for Medicare & Medicaid Services' meaningful use definition requires exchange capability for electronic health records (EHRs), we know very little about healthcare providers' motivations and employment of HIE systems.16–18 While many technological, political, and organizational barriers prevent the establishment of HIE, understanding why individual healthcare professionals actually utilize HIE is critical to ensuring its success and promoting further acceptance. This study, guided by information-seeking theory, addresses this current shortcoming by exploring the factors that are associated with actual HIE usage.
The factors associated with the usage of an HIE system to seek a patient's information can be modeled, in part, using constructs from Afifi and Weiner's theory of motivated information management.19 Here, HIE usage is posited to be a function of factors indicative of a need for information (interpretation) and an assessment of the potential value and opportunity costs of associated with information system use (evaluation). In the case of the interpretation, the discrepancy between an individual's actual and desired levels of uncertainty in a given situation generates anxiety. In turn, anxiety leads to a perceived need for information. Under the theory of motivated information management, interpretation is an assessment of whether or not the level of uncertainty associated with the task at hand warrants information seeking. This generalized view of uncertainty as a driver of information needs translates easily to the healthcare setting, as much of clinician information behavior is reflective of uncertainty about patient care.20–23 Although the availability of information from HIE is not a solution to all instances of uncertainty, two specific causes of uncertainty relate directly to the purpose of HIE. First, unfamiliarity is a source of uncertainty. For unfamiliar or unknown patients, providers rely on external medical documentation as an overview or memory aid.24 Since, during initial visits with new providers, relevant patient information is frequently incomplete or untimely,25–27 HIE system access during encounters with unfamiliar patients may reduce uncertainty.
Hypothesis 1: HIE usage will be more likely for encounters in which the patient is unfamiliar to the facility than for encounters in which the patient had previously visited the facility.
Second, task complexity due to either patient characteristics or previous utilization patterns also increases uncertainty. Specifically, the presence of multiple chronic conditions complicates care and requires more information to effectively manage,28 29 thereby creating additional uncertainty in treatment and diagnosis. Previous research revealed an association between HIE information system access and a patient's number of chronic conditions.11
Hypothesis 2: HIE usage will be more likely for encounters when patients have multiple chronic conditions.
Likewise, recent encounters with other providers complicates the provision of care, as treatments and diagnoses with one provider limit or constrain the available options to subsequent providers.30 31 For example, providers frequently do not receive patients' hospital discharge information32 or do not know about patients' recent emergency room usage.33 Therefore, patients who have frequent contact with the healthcare system not only may have ill health, but also may be among the greatest beneficiaries of improved information exchange.
Hypothesis 3a: HIE usage will be more likely for encounters in which the patient had frequent emergency room encounters in the past year than for encounters in which patients had not.
Hypothesis 3b: HIE usage will be more likely for encounters in which the patient had been hospitalized in the past year than for encounters in which patients had not.
Hypothesis 3c: HIE usage will be more likely for encounters in which the patient had frequent primary care encounters in the past year than for encounters in which patients had not.
The idea of evaluation requires an appraisal of both the expected benefits as well as the costs of information seeking.19 Since all information-seeking activities incur either resource or opportunity costs,34 individuals decide whether the time and effort to seek more information are warranted, or better served in another capacity.35 In the healthcare setting, previous research documents that physicians perform similar evaluations of their options and do not engage in information seeking in the face of time constraints.36 Likewise, nurses' information seeking is constrained by time limitations.37 In instances where opportunity costs would be high, use of HIE would be expectedly lower.
Hypothesis 4: HIE usage will be less likely for encounters with greater time constraints within a facility.
Finally, individuals assess the applicability of potentially sought information to their current context.19 The potential additional information may or may not have relevancy for the problem at hand. Stated alternatively, while HIE provides access to previously inaccessible externally generated information, not every encounter requires that type of information. For example, the main advantage of HIE appears to be access to diagnostic tests, existing treatments, and previous diagnoses.10 However, the treatment of select patient conditions does not rely heavily on this type of information. Instances indicating little value from externally created information may lead providers to avoid seeking addition information.38
Hypothesis 5: HIE usage will be less likely for encounters associated with injuries or accidents.
This study utilized a patient-level clinical dataset derived from the Integrated Care Collaboration (ICC) of Central Texas, a fully operational HIE effort encompassing Austin area safety-net providers. The ICC formed in 1997 as a non-profit organization with initial funding through federal and private grants. The 26 member organizations contribute annual membership dues scaled to their size and resources. ICC members include multihospital systems, public and private clinics, governmental agencies operating federally qualified health centers, and public health agencies. Similar to select efforts across the country, the ICC is an HIE effort serving the medically indigent. ICC does not systematically include encounters covered by private insurance or Medicare.
Member organizations contribute patient-level clinical and demographic data to a master patient index and centralized clinical data repository, I-Care, through secure electronic interfaces. I-Care is a proprietary system that exists independent of each organization's clinical data repository. Nightly extractions from the members' clinical data repositories concerning the data necessary for patient identification and matching, demographics, payors, encounter locations and dates, providers, diagnoses codes (ICD-9), procedure codes (CPT or ICD-9 depending on the member), and medications (if available) are uploaded to I-Care. By record matching in the master patient index through patient identifiers (along with periodic manual review by ICC system administrators) a patient record is compiled. In turn, authorized users at participating healthcare organizations may access the database via a secured website. Users include physicians, nurses, physician assistants, administrative staff, public health professionals, social workers, psychiatrists, and others; users receive introductory system training by the ICC staff on use of the I-Care system.
In addition, the ICC operates an opt-in model of HIE participation. Patients consent to system inclusion for a 2-year period by signing an authorization form at any point of service. This study only included patients who had previously consented that data about their individual healthcare encounters could be included in the I-Care data repository. Of the emergency room encounters eligible for study inclusion, 85.2% had authorized I-Care access. The study sample was drawn from all emergency department (ED) encounters among patients ages 18 to 64 that occurred between January 1, 2006 and June 30, 2009. We also excluded any emergency encounters occurring at facilities before the hospital had an authorized user of the I-Care system. The final dataset included 271 305 encounters (111 482 unique patients) from 10 facilities.
Users query and view patient records in I-Care through a series of specialized webpages (‘screens’). As part of the Health Insurance Portability & Accountability Act of 1996 (HIPPA) compliance, I-Care generates electronic logs to document users' activities (patient viewed, date, time, and screen(s) viewed). We transformed these logs into patterns of screens viewed for each user session. In the entire I-Care system, most user sessions (72.8%) were limited to the examination of only two screens—one enabling patient selection and another summarizing recent encounters. We classified all patterns that only included these two screens as basic usage, and the remaining sessions as novel usage. Novel patterns were marked by a greater variety of screens viewed, which included screens dedicated to patient contact information, payor histories, medication summaries or more detailed encounter records. This yielded a three-level multinomial outcome of (1) no usage, (2) basic usage, and (3) novel usage. Log files provide an objective measure of system usage unbiased by subject recall39 and are recommended for evaluating HIE.17 40
Since the system can be accessed by authorized users at any time, the user log is effectively encounter-independent. To link the dependent variable to patient encounters, we matched based on the master patient index unique patient identifier, date, user's reported work location, and place of encounter. Because late night ED encounters may actually trigger usage on the next calendar day, we also considered any user sessions beginning between midnight and 03:00 h as matching the day before. This method of matching resulted in two different forms of duplication that required specific attention. (These two forms of duplication were rare: the first case only occurred 69 times, and the second type 227 times in the final dataset; to make sure these did not bias the results, the analyses were conducted with and without these encounters; no substantive differences in results existed, so the observations were retained.) First, under our matching strategy, more than one user may access the system for a patient's record at a given encounter. In those instances where an encounter had more than one associated user session, if any of those sessions was a novel usage then we classified the usage as novel. The second form of duplication occurred when a patient visited the same ED more than once in the same day. We allowed those multiple encounters to link to a single user session based on two premises: (1) that it was impossible to determine which encounter was actually associated with the user session, so assigning any would be at best random error or at worst systematic error; and (2) if the user session occurred in response to the first encounter, the information obtained from HIE would potentially be available for the second visit. Lastly, encounters without an associated user session were classified as no usage, and this served as the reference category.
We constructed variables representing interpretation, evaluation, and control factors from data in I-Care. Unfamiliar patients were marked by the absence of any encounters at the same ED in the previous 12 months. We measured patient complexity in terms of the adapted Charlson comorbidity index score (excluding use of warfarin).41 The substance abuse and psychoses conditions from Elixhauser and colleagues' comorbidity list were included as independent predictors.42 We considered any diagnosis of these conditions at any type of healthcare encounter during the study period as indicative of having the condition. For prior utilization, we determined the total number of ED encounters, inpatient hospitalizations, and primary care clinic visits in the 12 months prior to the encounter date. For the number of ED encounters, we did not include visits at the same facility to avoid collinearity with the measure of patient familiarity. Following previous conceptualizations of encounter frequency, we divided ED and primary care visits into zero encounters, infrequent users (1–3), and frequent users (four or more).43–45 Because of small cell counts, inpatient hospitalizations were considered as a binary variable. Lastly, time constraint was defined as the ratio of total number of same-day encounters divided by the facility's previous year's mean number of daily encounters (by day of the week and month). We dichotomized this measure as busier-than-average day (daily encounters/historical average >1) or not a busy day. We used the Agency for Healthcare Research and Quality's Chronic Condition Indicator definitions to identify encounters associated with injuries and pregnancies and to describe the primary diagnosis. Because these are broad categories, including many different diagnoses, we present only descriptive measures (included in the, see online appendix). Lastly, we grouped the payer associated with the encounter into groups reflecting reimbursement rates: Medicaid, charity care & self-pay and multiple/no payers reported.
The unit of analysis was a healthcare encounter in an ED with the category of HIE usage (no use, basic or novel) as the outcome. Frequencies and percentages described the study sample. Hypotheses were examined using crossed random-intercept logistic regression models,46 which account for the clustering of encounters within patients and patients within locations, and allow patients to have encounters at different facilities. Since the reference category is substantially larger than the outcomes of interest, independent binary logistic equations were fit instead of a multinomial logistic regression,47 considering all variables discussed earlier as fixed effects. To adjust for confounding, we created best-fitting multivariate models in Stata using a backward selection modeling approach looking for improvements in Akaike and Bayesian Information Criterion values. All variables hypothesized to be associated with information seeking were retained during the modeling building process. For hypothesis testing, we set α=0.05. However, because of the testing of multiple hypotheses in the multivariate models, we applied the Holm correction to the calculated p values.48 The parameter coefficients were exponentiated to express ORs.
System users accessed the I-Care system for 2.3% of the 271 305 encounters included in the study (table 1). This level of usage is consistent with existing reports in the literature.13 14 Basic usage (n=2527) accounted for 41.1% of instances. In contrast to emergency encounters nationwide, the sample was predominately Hispanic, younger, and a higher proportion of charity care recipients (owing to the location and characteristics of the HIE).49
Even though we only included visits after each facility had at least one authorized users of the HIE system, much variation in usage occurred across facilities. One of the 10 facilities used the system for nearly 4% of patients, while half did not make use of the system at all. Additionally, we noticed a degradation of usage over time.
Table 2 contains the unadjusted and adjusted ORs for basic usage compared to no usage. After adjusting for confounding, nine factors were significantly associated with basic usage according to corrected p values. As hypothesized, the odds of basic HIE usage were lower when time constraints increased. On days with a higher number of encounters than average, the odds of basic usage were 17% lower. Second, the hypothesis that usage would be lower for situations in which less external information may be necessary was not completely supported. After correcting for multiple hypothesis testing, injury-related encounters were not associated with basic usage (OR 0.89; 95% CI 0.79 to 1.00). However, encounters due to alcoholism were negatively associated with basic usage and may be situations where the delivery of emergency care does not depend greatly on information collected in other organizations. The hypothesized relationship between complexity and usage was partially supported. Prior primary care clinic usage was also associated with usage (OR 1.42; 95% CI 1.27 to 1.58 and OR 1.37; 95% CI 1.22 to 1.54). However, unexpectedly, neither ED usage at other facilities, recent hospitalizations, nor comorbidity were associated with basic usage after adjustment. Two non-clinical factors were associated with usage as well. Charity care was positively associated with basic usage, and the odds of usage were lower for African–Americans, Hispanics, and patients of other race/ethnicities. The observed relationship between patient familiarity and basic usage contradicted the relationship that we hypothesized. Unfamiliar patients had significantly lower odds of basic usage (OR 0.34; 95% CI 0.30 to 0.37).
Also presented in table 2 are the adjusted and unadjusted associations for novel usage; the statistically significant variance of the intercepts of the adjusted model indicates the appropriateness of the random-effects model. As in the case of basic usage, odds of novel usage were lower for encounters on busy days (OR 0.84; 95% CI 0.78 to 0.90) and with unfamiliar patients (OR 0.32; 95% CI 0.29 to 0.35). Likewise, frequent primary care visits (OR 1.76; 95% CI 1.60 to 1.94) and charity care encounters (OR 1.51; 95% CI 1.33 to 1.72) were each positively associated with novel usage. However, novel usage was different in respect to cormorbidities and prior hospitalizations. As hypothesized, higher scores on the categorized Charlson index were positively associated with novel usage (OR 1.19; 95% CI 1.07 to 1.32 and OR 1.18; 95% CI 1.07 to 1.30) after adjustment. The odds of novel usage were also significantly higher in cases of prior hospitalization (OR 1.34; 95% CI 1.22 to 1.48).
Implementation of HIE by an organization does not ensure utilization by individuals within the organization. Research has repeatedly demonstrated the organizational decision to adopt an innovation is frequently independent of individuals' adoption decisions.50 51 This exploratory study supports the point in the context of voluntary use HIE, but also suggests when and why such systems are actually used and how to improve implementation.
First, these results identify time constraints as a barrier to HIE usage. This simple finding, consistent with information-seeking theory and prior research, has immediate application to the design and function of HIE efforts. Healthcare is busy and fast-paced, and some physicians already believe HIE may not save time.52–54 Given that the voluntary access of an additional information source can be discouraged by time constraints, those wishing to implement HIE have two clear options. One avenue is to improve the utility of the information and the system; in effect, change the equation so the potentially available information is more valuable than the opportunity costs. Screen redesign, single sign-on, eliciting user needs, or improved record searching could all be means to that end. Alternatively, organizations can dramatically increase the level of functional integration between exchange partners' EHRs and their own. Case reports suggest tighter functional integration is associated with higher proportions of usage,12 and the problem of constrained time illustrates why. Directly placing the information made available by HIE into the organization's EHR may be politically, legally, organizationally, and even technically difficult, but it effectively removes from the user the decision to seek or not to seek information in such an alternative source.
However, these results provide an argument against addressing the problem of low usage by simply mandating usage. The decreased odds of usage for some encounters, but increased for others suggests users, have determined HIE is useful in some, but not all, instances. For current encounters that have little to do with previous utilization or information stored in other organizations, HIE may have less immediate value. Furthermore, the effect of comorbidities on novel usage, but not on basic usage, indicates users employ the system in a fashion to meet their immediate needs. For complex patients, the minimum information provided by the HIE system was probably not sufficient; those encounters required more detailed investigation. Mandating usage of an alternative information source that changes work processes needlessly, or when little potential value for the problem at hand exists, is a prescription for inciting resistance.55 Therefore, blanket mandatory usage of HIE systems may not solve perceived problems of underutilization.
Usage was much less likely for unfamiliar patients contradicting both expectations based on theory and conventional wisdom. An unknown patient is in effect the poster child for justifying HIE in the ED setting.56 57 Such a view is understandable, as repeated contact should increase provider knowledge about the patient's history and idiosyncrasies, thereby lessening the need to seek additional information. However, this unexpected relationship suggests one very practical reason why HIE, at least in the emergency setting, is actually used. In the ED, patient familiarity is undesirable, because it is indicative of patients with inappropriate sources of care. For the familiar patient, HIE might provide clinicians and organizations the necessary information to get and keep these patients out of the ED. The absence of an association between ED utilization at other locations and HIE usage reinforces this interpretation. Patient familiarity with a facility is more important than the frequency of a patient's visits to any ED.
Lastly, HIE usage appears to have non-clinical reasons as well. As noted, evidence suggests it is used in response to a facility's repeat patients, and the association between payer type and usage is also suggestive. While Medicaid does not boast the most generous reimbursement rates, some payment is better than no payment. Among patients for whom no payment could be expected, system usage was higher, suggesting the possibility of either using the system to locate past payers or to again help find patients more appropriate sources of care.
The primary limitation of this type of log file analysis is that we do not know if information-seeking efforts were successful. The files only record that a particular screen was viewed. We do not know if the user was able to locate the desired information on that screen or if it even existed. We simply know the user looked. Further studies with different designs would be necessary to investigate the success of search efforts.
Also, the results of this study have limited generalizability in terms of information system characteristics, the setting of care, and the population served. The HIE system considered in this study is a standalone system, separate from each organization's EHR. As the level of functional integration changes, the opportunities for users to access information from HIE systems also changes. In addition, the usage of separate information systems may require more substantial adjustments to work processes than are necessary to use systems with higher levels of integration. Other system or exchange effort characteristics such as technical architectures, user interfaces, and participating organizations may also limit generalizability. In addition, these ED-based finding may not be generalizable to the inpatient hospital setting, primary care encounters, or public health usage. Lastly, findings may not hold for HIEs serving broader payor groups as service needs and patterns of usage among the medically indigent in the ED may be substantially different from insured populations.
These results are limited in terms of scope, measurement, and causality. First, this study does not address any potential confounding due to user characteristics. While a necessary avenue of future inquiry, this study could not utilize measures of system users primarily because user activity is only recorded in the system for those encounters in which the system was accessed. Therefore, user characteristics could only be attributed to decisions to seek information, but not encounters where the system was not utilized. Second, the usage construct is limited because it does not include the specific information sought, particular search strategies, how usage fit into the patient encounter, or even if the search successfully yielded the desired information. Additionally, because our matching strategy was geared toward linking to patient encounters, we effectively excluded unsuccessful searches in our usages counts. Therefore, actual user interaction with the system was undoubtedly more frequent than our measure reported. Third, the strategy for linking user sessions to patient records was developed specifically for this study: it may not be applicable to other investigations. Furthermore, this method excluded usage by disease-management programs, social services, or public health. These constitute important applications not addressed in this investigation. Fourth, the cross-sectional design does not eliminate the potential bias from attrition. As noted, encounters covered by private insurance are not in the dataset. Therefore, this study may be undercounting previous utilization and previous diagnoses for some individuals. However, this limitation has no influence on the effect of a busy day at the ED (because that measure is patient-independent), injury (because it is unique to the encounter), or measures that are static across patients (ie, gender, race/ethnicity).
The meaningful usage criteria resulting from the provisions of the Health Information Technology for Economic & Clinical Health (HITECH) Act, part of the American Recovery & Reinvestment Act of 2009, will undoubtedly expand the number of organizations pursuing HIE. However, simply mandating the connections for exchange will not be enough. The presence of HIE does not translate to widespread use, individual usage can be hindered, and usage is not appropriate in all instances. As meaningful use definitions evolve in subsequent years, attention will have to be centered on how healthcare professionals actually employ HIE systems if there are to be any hopes of accruing the promised improvements in patient health and the delivery of care.
We would like to thank D Brown and A Khurshid, at the Integrated Care Collaborative of Central Texas, and S Coe Simmons, for their assistance in obtaining the data for this study.
Funding This work was supported by Award Number R21CA138605 from the National Cancer Institute.
Competing interests None.
Ethics approval The Office of Research Compliance, Texas A&M University.
Provenance and peer review Not commissioned; externally peer reviewed.