Different approaches to improve cohort identification using electronic health records: X-linked hypophosphatemia as an example

Summary

Electronic Health Records (EHRs) represent a source of high value data which is often underutilized because exploiting the information contained therein requires specialized techniques unavailable to the end user i.e. the physician or the investigator. Here I describe four simple and practical avenues that will allow the standard EHR end user to identify patient cohorts: the use of diagnostic codes from different international catalogues; a search in reports from complementary tests (e.g. radiographs or lab tests) for any result of interest; a free text search; or a drug prescription search in the patient's electronic prescription record. This medical approach is acquiring great importance in the field of rare diseases, and here I demonstrate its application with X-linked hypophosphatemia. The use of these four EHR questioning approaches makes finding a cohort of patients of any condition or disease feasible and manageable, and once each case record is checked, a well-defined cohort can be assembled.

1. Introduction

Electronic Health Records (EHRs), of common use today, have changed the way clinical information is acquired and documented. They provide quick and economic access to vast amounts of data for clinical and research purposes. However, the exploitation of this information normally uses technical methods, such as natural language processing (NLP), text mining or machine learning techniques which in turn require specialized bioinformatic processes for their implementation (1). As a result, this vast amount of data may be inaccessible to the end users, such as clinicians or investigators, who are often looking for patients to recruit for clinical trials or their own observational studies (2). The absence of end-user tools greatly diminishes the EHRs' potential as a source of high value information into an underutilized resource that is exacerbated by the number of electronic health record software developers (3) and the lack of integration between them (4).

In recent years interest has especially grown in the area of rare diseases, which involves thousands of very different diseases that affect a small number of people within the general population (5). In Europe, rare diseases are defined as those with a prevalence of less than 1 in every 2,000 people, and within this group a disease is considered to be ultra-rare if it affects 1 in every 50,000 people (6). Despite the rising awareness of rare diseases, they still represent a challenge for clinicians, investigators, and public health systems (7). The small number of affected patients makes it challenging to investigate the pathophysiological mechanisms and discourages many companies from investing in treatments. In addition, few doctors recognize the heterogeneous signs and symptoms, leading to missed or delayed diagnoses. The length of time for the diagnosis of a rare disease is around 4.8 years, with the patient seeing an average of 7.3 different specialists for case evaluations (8). This delay has a serious impact on the patient's quality of life. Earlier identification of the disease is vital, not only for patient follow-up but also for furthering research.

Here, I describe different but compatible approaches that make it possible for a patient cohort to be identified by the standard EHR end user. Whether all or some of these approaches are workable will depend on the EHR developer and the information access provided by the software. For our example, I will apply the four different strategies to the rare disease, X-linked hypophosphatemia.

2. X-linked hypophosphatemia

The estimated incidence and prevalence of X-linked hypophosphatemia (XLH) are less than 1 in 2,500 and 1 in 20,000, respectively (9). Despite the prevalence of this disorder being similar to other known pathologies, and the existence of specialists involved in its diagnosis and treatment, awareness and knowledge about this disease are still relatively low.

XLH has its origin in the mutation of the PHEX gene, with a dominant X-linked inherited trait. The mutated PHEX translates into an increased serum concentration of fibroblast growth factor 23 (FGF-23), a hormone whose main function is to regulate serum phosphate levels. The decreased clearance of FGF-23 leads to decreased tubular phosphate reabsorption which, together with decreased intestinal absorption of phosphorus, induces hypophosphatemia and ultimately the clinical manifestations of the disease (10,11).

The signs and symptoms, which can vary in severity, are evident from the first months of life. The most typical clinical features, which are soon noticeable in childhood, are bowed legs (genu varum) and short stature. But there may be other manifestations such as knock-knee (genu valgum), frequent fractures, osteoarticular pain, extraosseous calcifications, dental problems or hearing impairment, which in many cases cause a significant functional limitation that negatively impacts the patient's quality of life (12). Once suspicion of XLH is raised from the physical examination or the imaging tests, the diagnostic approach should begin by ruling out other causes of rickets, especially deficiencies. Patients with XLH have normal parathyroid hormone (PTH), vitamin D, and calcium in the blood and urine, but serum phosphate levels and tubular phosphate reabsorption are low (11).

Conventional treatment is based on oral phosphorus supplements and active vitamin D analogues to maintain alkaline phosphatase within normal range and minimize bone deformities, therefore early diagnosis is essential (13). However, these treatments are not without adverse effects, of which the most common are gastrointestinal disorders and the most serious are secondary hyperparathyroidism and nephrocalcinosis (14). The development of a new monoclonal antibody has raised new hopes in the management of the disease (15,16), as it has shown to stabilize serum phosphate levels and lead to improvements in rickets, skeletal healing, and physical function (13). Doctors need to take into account all their current and past patients who could benefit from any therapeutic breakthrough, and those who could participate in new clinical trials and observational studies. This is only possible if XLH patients can be correctly identified.

3. Methods for cohort identification

I demonstrate an EHR search of XLH patients with four easy-to-use data sources.

3.1. Diagnostic codes

One search method available in virtually all institutional systems is searching the diagnostic codes of international catalogues. However, a frequently encountered difficulty is that most catalogues group diseases together with other entities, so extensive case information is needed for correct screening, especially when dealing with rare diseases. This can be achieved using the other methods described below.

The first option is the widely used International Statistical Classification of Diseases and Related Health Problems (ICD) with its different editions: ICD-9, -10 and -11 are the editions most commonly used in hospitals. Many studies have been published using this classification for patient identification (17-19).

In ICD-9, XLH is considered a disorder of phosphorus metabolism and the name given is old terminology: vitamin D-resistant rickets, code 275.3. ICD-10 goes a step further in the classification of these disorders of phosphorus metabolism by a separation into four groups: XLH is here coded as E83.30, which encompasses unspecified disorders of phosphorus metabolism. However, ICD-11 made a major change in the classification. It removed XLH from the section of phosphorus metabolism disorders and placed it in the Disorders of vitamin D metabolism or transport, forming its own group, hypophosphatemic rickets, encoded as 5C63.22.

Another widely used classification tool is the Diagnosis-Related Group (DRG), which attempts to classify the diagnosis as reimbursable "products" provided by the hospital, which means that it is designed more for billing than for clinical data analysis (20). This system normally places rare diseases, such as XLH, in a general category, classified under "Inborn and other disorders of metabolism", coded as 642.

On the opposite end of the scale is the Systematized Nomenclature of Medicine - Clinical Terms (SNOMED-CT) which is a coded, comprehensive, multilingual, clinical terminology that is becoming a standard in many countries (21). Here XLH is given a unique code: 82236004. Another classification database available is the Online Mendelian Inheritance in Man (OMIM), which catalogues all known human genetic disorders; XLH as a hereditary disease has its code: 307800. The summary of the codes for XLH is shown in Figure 1.

Figure 1.

Diagnostic codes for X-linked hypophosphatemia. DRG, diagnosis-related group; ICD, International Statistical Classification of Diseases and Related Health Problems; OMIM, Online Mendelian Inheritance in Man; SNOMED, Systematized Nomenclature of Medicine.

3.2. Complementary tests

Another way to find a cohort of patients among the big data provided by the EHRs is to look for diagnostic criteria already discussed in the lab test results or in the imaging reports. Once these findings have been filtered, each case must be reviewed to generate a differential diagnosis that will confirm or rule out the disease.

In the case of XLH, the most effective strategy would be to filter cases of binomial decrease in tubular phosphate reabsorption and hypophosphatemia. This could raise suspicions of phosphopenic rickets. Even so, we would still have to rule out calcipenic rickets (also known as vitamin D-deficient rickets) by focusing on PTH and vitamin D levels, and Fanconi's syndrome, by looking at urine tests positive for glucosuria, high pH levels and elevated fractional excretion of potassium (22). Other approaches that follow this strategy would be to look for the typical radiographic lesions (metaphyseal cupping and flaring and physeal widening and irregularity of the long bones). Additionally, a search for PHEX gene mutations should be made in the genetic results database (Figure 2).

Figure 2.

Complementary tests approach for X-linked hypophosphatemia.

3.3. Free text search

In most EHRs there is the possibility for the clinician to both code their findings in a structured format and also enter information in narrative free text. Unstructured free text searches, when available, can be more efficient than searching for a diagnostic code (23). There are different ways to search for free text. One way is to perform a plain text search by entering the different names given to the same disease. In the case of XLH, the different terms include: X-linked hypophosphatemic rickets, X-linked hypophosphatemia, vitamin D-resistant rickets, etc. Another alternative would be to search for text in clinical notes, imaging or genetic test reports. This strategy is more far-reaching but less specific than searching for diagnostic codes. It may be time-consuming for very prevalent diseases, but in the case of rare diseases it is best to begin with a long list of potential patients (24). Figure 3 shows an example of a free text search for X-linked hypophosphatemia.

Figure 3.

Free text search approach.

An alternative or complementary approach is the use of natural language processing (NLP) to extract important information from text-based documents. This tool has proven to be more accurate in identifying postoperative complications compared with discharge coding information (25). It has also shown advantages in the classification of results of brain tumour magnetic resonance imaging reports in comparison to human expert classification, with excellent accuracy observed for tumour status classification (26). Unfortunately, free text searches are rarely exploited (27,28).

3.4. Electronic prescriptions

Finally, another way to find cohorts is to search for prescriptions in the patient's electronic medical records. Some drugs indicate specific pathologies, but if that is not the case for the disease being queried, the combination of drugs with several indications may sometimes lead to the disease in question. For example, this method is especially useful in XLH since the classical medical treatment is based on oral supplementation of phosphorus and vitamin D. These two medicinal products, often used separately, are prescribed together in very few situations (Figure 4). This approach can be a useful screening step to then be complemented by the review of each case to confirm the identification. It can be especially valid for those adult patients who received treatment in childhood and were subsequently lost to follow-up (29).

Figure 4.

Patients' drug prescription approach.

4. Conclusion

The identification of cohorts of patients with a certain condition or pathology using "EHR big data" is a growing concern (1,30). Sadly, this has become overly technical and out of the reach of doctors or investigators who are not experts in these new technologies (31). Therefore, for the EHR data interrogation to successfully satisfy the end user, easier ways are needed to obtain and analyse the information. I have introduced four different ways of finding patients that may be available to the entire medical community. I am aware that each institution with its own EHR software may not have all four of the proposed methods available. However, I am confident that just one or a combination of the methods would facilitate the task of finding patients with a particular condition. In the case of a rare disease, such as XLH, it is preferable to rely on more sensitive methods, as an eventual "false positive" can always be discarded afterwards. With a well-defined cohort, any investigation or clinical trial can be launched (32).

Funding: This article has been commissioned and financially supported by Kyowa Kirin. Neither the author nor the founder have economic or patrimonial interests in the methodological approach explained in this article. The author states that the funder was not involved in the preparation of the manuscript. The author wishes to thank Anabel Herrero, PhD for providing editing assistance. Kyowa Kirin funded this assistance.

Conflict of Interest

JJB reports personal fees and non-financial support from Kyowa Kirin during the conduct of the study.