Determinants of frequency and longevity of hospital encounters' data use

The age of data is one of the factors usually used to assess data relevance, typically making new information more relevant to the current search. As an example, data that were at least three days old has been categorized as "old data" in the emergency setting [1]. We argue that data ages are likely to differ according to its type, i.e., some clinical reports are less ephemeral than others and so are found useful longer than others. Also, the context of healthcare (e.g.: hospital environment, primary care, oncology), health conditions and patient age probably influences the way information maintains its relevance [2].

In this study we aimed to determine for how long clinical documents are used by health professionals in a Hospital Information System considering the setting of information request and its content.

The practice of medicine has been described as being dominated by how well information is collected, processed, retrieved, and communicated [3]. Patient records, the patient and published evidence are the three information sources needed to practice evidence-based medicine [4], being complemented with the health professional own experience in daily routine.

There is a great demand to create efficient integrated electronic patient records to facilitate the communication process between health professionals by delivering in a single interface data from many different information systems [7]. These systems are evolving to meet people's needs by implementing larger networks, allowing patient access and integration of ever more items of patient data [8]. Although great advances have been made over the years, on-demand access to clinical information is still inadequate in many settings, contributing to duplication of effort, excess costs, adverse events, reduced efficiency [9] and inability to take full advantage of existing IS [10]. Although it is widely accepted that full access to integrated electronic patient records and instant access to up-to-date medical knowledge significantly reduces faulty decision making resulting from lack of information [5, 11, 12], there is still very little evidence that life-long EHR improve patient care [13]. Shapiro et al. found that, although emergency department doctors believe their patients would benefit from longitudinal records, they only try to obtain such data in 10% of the cases [14]. Recently Hripcsak et al. [1], described the access rates to WebCIS in the emergency department. The results indicate that data generated before the current emergency visit are accessed often, but not a majority of the times, even when the user is notified of the availability of data.

Jones showed that viewing lab results within the context of the patient record made physicians alter their ordering patterns accordingly [15]. Distinguishing between relevant and useless information enables enhanced graphical user interfaces (GUI) to highlight the most relevant information, as well as improved data management by choosing storage devices with better performance for relevant data. However, it is hard to understand what information is really important to daily clinical care, and what is simply occasionally used [16].

Between May 2003 and May 2004, the Department of Biostatistics and Medical Informatics implemented a Virtual Patient Record (VPR) [17] for the Hospital S. João (HSJ), a tertiary university hospital with over 1,350 beds that is the main referring hospital in the north region of Portugal for all medical and surgical specialties.

The system integrates clinical data from 12 legacy Departmental Information Systems (DIS) and the Diagnosis Related Groups and Hospital Administrative databases, aiming to deliver the maximum information possible to health professionals. Over 1,300 medical doctors use the system on a daily basis and the HSJ-VEPR retrieves an average of 3,000 new reports each day [17, 18], adding up to more than 3 million reports collected so far.

The main components of the VPR are a web-interface (see Figure 1), an integration system and a central repository. The web-interface was designed to include graphical components and layouts to summarise past patient data (patient chronological bars), and folders that reproduce the traditional types of patient record organisations (source, chronological and problem views). It allows ubiquitous access to heterogeneous data sources. The VPR was made available for testing in October 2004, and since December 2004 it started to be known and routinely used.

The authors of this paper have done some preliminary studies aiming to answer for how long are clinical documents useful for health professionals in a hospital environment considering its' content and the context of information request. Our results show that some clinical reports are still used after one year regardless of the context in which they were created, although significant differences exist in reports created in distinct encounter types. The median life of reports by encounter type is 1.7 days for emergency, 3.9 days for inpatient and 27.7 for outpatient encounters. We concluded that the usage of patients past information (data from previous hospital encounters), varied significantly according to the setting of healthcare and content [19].

In this paper we extend our previous work to include the study of associations between the visualization of clinical reports and patient characteristics (e.g. age), the source of the reports, the type of hospital encounter and the main diagnosis for the hospitalization. Also, a mathematical model on how the reports are seen over time is explored.

Although the median life of both genders is very similar (20.09 and 20.11 days for men and women), 25% percent of the men reports are seen in the first day whilst it takes two days for women reports to get to the same percentage. Reports tend to have higher median lives as the patients' age is nearer of middle age (40 years). These results are probably related to the most common diseases in each gender and age categories and the period patients are followed in hospital encounters.

Report median lives are very different depending on the departmental source and report content. They range from 1 day of haemostasis reports (percentile 75 is 13 days) to 118 days of pathology reports (percentile 75 is 346 days). These results clearly show that different reports are needed faster (e.g. percentile 25 of haemostasis is 0.1 days ≈ 2 hours) than others (e.g. percentile 25 of gynaecology is 35 days). It also shows that some reports are needed for longer periods of time (e.g. percentile 75 of pathology is 346) than others (e.g. percentile 75 of haemostasis is 13 days).

The analysis of report median lives according to ICD-9-CM shows a close relation between report usefulness along the time and the main diagnosis related to each inpatient encounter; the median life of neoplasms (median life = 36 days; p25 = 1.8; p75 = 268) is 51 times higher than injury and poisoning (median life = 0.7 days; p25 = 0.1; p75 = 6). The main diagnosis of the inpatient encounter certainly is associated with the possibility of happening future encounters within the same hospital (e.g. chronic diseases and neoplasms are followed for longer periods than injuries or poisoning). Unfortunately in this scenario, it is not available a similar diagnosis code describing the reason for encounter in outpatient, emergency and day-care encounters.

Two distinct phases can be defined in inpatient and emergency encounters: firstly an initial patient observation and evaluation based on past and current information help to achieve a diagnosis; secondly follows a sequence of iterations of the diagnostic-therapeutic cycles in which after each therapeutic action new observations are made (e.g. lab analysis) to monitor the patient state [20]. The main difference between both phases is that in the first phase there is a more active use of information collected in previous encounters. The longer the encounter (e.g. inpatient are normally longer than emergency encounters) more iterations occur in the second phase resulting in more views of concomitant encounter reports (64% in inpatient versus 47% in emergency) than previous reports.

However, the large number of views in emergency encounters referring to old reports (53% of all report views) should be stressed as these results probably show a higher use of past data in our case, than on other published studies (5% to 20% of all emergency encounters in Hripcsak et al. [1], and 10% of all emergency encounters in Shapiro et al. [14]). Moreover, these reports were initially produced in different encounter types, which illustrate the importance of longitudinal electronic health records even in emergency scenarios.

It is important to note that most of these reports when viewed in a new encounter, gain a second life after their initially planned life cycle that was limited to the encounter where it was produced. This fact illustrates how hard is to define what information is really important to clinical care, and what is simply occasionally useful. In the outpatient scenario, one could expect report views to show more clearly a long-term problem-driven patient healthcare. But the number of views if previous emergency and inpatient encounter, shows the importance to medical specialty encounters of unplanned past events.

Regarding the mathematical model, the high values of R² (> 0.9) shows that the exponential regression is a good fit to model how the reports are seen over time. This model can be used to estimate the relevancy of each report in a particular moment.

Overall, our results show that in the VPR system past patient information is used by doctors in everyday healthcare. Like in Hripscsak et al. this proves that when doctors have access to past information, even if it was collected for a different purpose, they use it to make a diagnosis [1].

Future work

The availability of patient information in integrated clinical Information Systems raises a new kind of problem - too much data, and too little time to select and read it. These systems are collecting hundreds of reports per patient and presenting them for the user to decide which he will select to read. Helping the health professional filter the reports may be the difference between finding relevant information or not. Report usefulness (taking into consideration their median life) varies between 7 and 118 days depending on the department that originated that information. We feel that these variations should have consequences on information presentation and management, instead of pre-determining when information gets old. As an example one could adjust the display time window for recent reports to the estimated median life of each report. Our results show that the context of the current encounter, the type of encounter and department where the report was produced are related to the report longevity.

These results will be influenced by how a particular Hospital operates, how the health professionals work, what IS are made available, and if there are paper records at hand. Each IS should carry out a similar analysis of report use logs, and adapt its GUI and data management to it. If a group of reports of Clinical Pathology were produced before the last 15 days, their relevance should probably be downgraded, leaving space for other reports. On the other hand, pathology reports should be kept online in fast digital archives for much longer, due to the fact that many are still accessed after 18 months years.

Figure 4 shows a prototype GUI that enhances clinical report usefulness by taking into consideration the calculated median life of each report by department. This interface would promote some reports by others according to a calculated time interval for each department, in this case set to the median life (median of previous viewings of same type of reports) of the reports. The authors are implementing interfaces similar to the proposed one and evaluating its usability by health professionals.

The usage of past patient information in the VPR case-study varies significantly according to patient age, type of information, type of hospital encounter and medical cause (main diagnosis) for the encounter. As more and more patient information is stored, it is very important to efficiently select which one is more likely to be useful and promote it in a scenario where scarcity of resources (screen space, storage space, bandwidth and doctors' time) is very real.

Researchers and developers dealing with the implementation, monitoring and utilization of an integrated Hospital Information System should create mechanisms that automatically take in consideration the usage profiles to efficiently manage available resources.