This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Medical Informatics, is properly cited. The complete bibliographic information, a link to the original publication on http://medinform.jmir.org/, as well as this copyright and license information must be included.
Direct feedback on quality of care is one of the key features of a learning health care system (LHS), enabling health care professionals to improve upon the routine clinical care of their patients during practice.
This study aimed to evaluate the potential of routine care data extracted from electronic health records (EHRs) in order to obtain reliable information on low-density lipoprotein cholesterol (LDL-c) management in cardiovascular disease (CVD) patients referred to a tertiary care center.
We extracted all LDL-c measurements from the EHRs of patients with a history of CVD referred to the University Medical Center Utrecht. We assessed LDL-c target attainment at the time of referral and per year. In patients with multiple measurements, we analyzed LDL-c trajectories, truncated at 6 follow-up measurements. Lastly, we performed a logistic regression analysis to investigate factors associated with improvement of LDL-c at the next measurement.
Between February 2003 and December 2017, 250,749 LDL-c measurements were taken from 95,795 patients, of whom 23,932 had a history of CVD. At the time of referral, 51% of patients had not reached their LDL-c target. A large proportion of patients (55%) had no follow-up LDL-c measurements. Most of the patients with repeated measurements showed no change in LDL-c levels over time: the transition probability to remain in the same category was up to 0.84. Sequence clustering analysis showed more women (odds ratio 1.18, 95% CI 1.07-1.10) in the cluster with both most measurements off target and the most LDL-c measurements furthest from the target. Timing of drug prescription was difficult to determine from our data, limiting the interpretation of results regarding medication management.
Routine care data can be used to provide feedback on quality of care, such as LDL-c target attainment. These routine care data show high off-target prevalence and little change in LDL-c over time. Registrations of diagnosis; follow-up trajectory, including primary and secondary care; and medication use need to be improved in order to enhance usability of the EHR system for adequate feedback.
At present, quality of care is generally evaluated in clinical trials or in expensive and laborious cross-sectional studies, such as the European Action on Secondary and Primary Prevention by Intervention to Reduce Events (EUROASPIRE) or SUrvey of Risk Factor management (SURF) initiatives, which evaluated target attainment of low-density lipoprotein cholesterol (LDL-c) [
Cardiovascular risk management (CVRM) is an example for complex care, with many factors and physicians involved over a long period, that could benefit from an LHS approach. Risk-factor level reduction and control is key in primary and secondary cardiovascular risk prevention. In particular, pharmacological LDL-c-lowering treatment is one of the cornerstones of cardiovascular disease (CVD) prevention, leading to a large risk reduction [
In this study, we evaluated the potential of routine clinical care data extracted from electronic health records (EHRs) to obtain reliable information on LDL-c management in CVD patients referred to a tertiary care center.
We conducted a prospective study with data extracted from the EHRs of patients of the University Medical Center (UMC) Utrecht, Utrecht, the Netherlands. All data from the EHRs of the UMC Utrecht are stored in the Utrecht Patient-Oriented Database (UPOD). In short, this database comprises all clinical information, demographic data, medication, diagnoses, and lab measurements, directly extracted from the EHRs of patients who visited the UMC Utrecht from 2003 onward, encompassing data from more than 2 million individual patients to date [
All patients with at least one documented LDL-c measurement in the database were included in the study. This study’s analysis was restricted to patients with established CVD, as these patients have an indication for LDL-c management according to the Dutch guidelines [
All LDL-c measurements in adult patients (≥18 years of age) available at the UMC Utrecht were retrieved from the UPOD. In patients for whom all other lipids but LDL-c were measured, LDL-c was calculated using the Friedewald formula [
We extracted information on sex, age, diabetes mellitus, hypertension, chronic kidney disease (CKD), blood pressure, smoking status, and use of blood pressure-lowering, lipid-lowering, or blood glucose-lowering medication. Sex and age were extracted from the general hospital administration data, which are checked via identification during the first visit at our center. History of diabetes mellitus was based on diagnosis codes, financial billing codes, and prescription of blood glucose-lowering medication. Hypertension was defined as blood pressure over 140/90 mmHg and/or prescription of blood pressure-lowering medication. CKD was defined using diagnose codes; interventions, including dialysis and shunt surgery; or estimated glomerular filtration rate levels that were extracted from the laboratory system within 48 hours around the LDL-c measurement. Smoking status was retrieved from predefined tables, dedicated to smoking registration, as well as from free text. Blood pressure-lowering, lipid-lowering, blood glucose-lowering, and antithrombotic medication data were extracted from the electronic prescription system using the Anatomical Therapeutic Chemical classification codes starting with A10, B01, B02A, and C02-C10. We converted statin dosages to atorvastatin 20 mg equivalent dosages (see Table MA1-1 in
After extracting LDL-c measurements from the database, we excluded patients with unreliable LDL-c values, as described above, and patients without established CVD. We divided the remaining group into patients with repeated measurements and patients without repeated measurements.
First, we calculated the prevalence of target attainment at the first measurement per patient, which was the only measurement for the patients without repeated measurements. The LDL-c target was defined as less than 2.5 mmol/L, according to the Dutch CVRM guideline [
Second, we investigated the trajectories of LDL-c distributions in patients with repetitive measurements. For the repetitive measurements, we distinguished different follow-up scenarios (see
Visualization of possible follow-up scenarios.
Lastly, we assessed factors associated with unfavorable LDL-c category change. Favorable change was defined as an LDL-c decreasing to or remaining on target. Unfavorable change was defined as an increase in LDL-c, a decrease in LDL-c but still off target, or a stable LDL-c that was off target. We performed a logistic regression analysis with deterioration as the outcome and age, sex, diabetes, hypertension, smoking, antithrombotic agent use, statin change (type and dose), the number of the measurement, and follow-up time (short- or long-term) as covariates.
All analyses were performed in R statistical software, version 4.3 (The R Foundation).
A total of 250,749 LDL-c measurements were collected from 95,795 individual patients at the UMC Utrecht between February 2003 and December 2017 (see
Flowchart of data retrieval for the study. CVD: cardiovascular disease; LDL-c: low-density lipoprotein cholesterol.
In 23,932 patients with CVD, LDL-c was measured repeatedly in 10,771 patients (45.00%) and once in 13,161 patients (54.99%) (see
The distributions of LDL-c categories (see
Baseline characteristics for cardiovascular disease (CVD) patients at first measurement in strata of presence of repeated measurements.
Characteristic | No repeated measurements (N=13,161) | Repeated measurements (N=10,771) | |
Women, n (%) | 4257 (32.35) | 3254 (30.21) | |
Age (years), mean (SD) | 65.5 (12.8) | 60.8 (12.1) | |
Smoking (current), n (%) | 1523 (11.57) | 967 (8.98) | |
LDL-ca (mmol/L), median (IQR) | 2.4 (1.9-3.1) | 2.4 (1.9-3.1) | |
Systolic blood pressure (mmHg), mean (SD) | 137.5 (23.5) | 135.3 (23.2) | |
Diastolic blood pressure (mmHg), mean (SD) | 76.3 (13.5) | 77.5 (13.7) | |
Diabetes, n (%) | 1456 (11.06) | 1415 (13.14) | |
Hypertension, n (%) | 4428 (33.64) | 3514 (32.62) | |
Chronic kidney disease, n (%) | 43 (0.33) | 108 (1.00) | |
|
|
|
|
|
Coronary heart disease | 9313 (70.76) | 7660 (71.11) |
|
Stroke | 2912 (22.13) | 1929 (17.91) |
|
Peripheral artery disease | 1461 (11.10) | 1791 (16.63) |
|
Abdominal aortic aneurysm | 502 (3.81) | 503 (4.67) |
|
|
|
|
|
Statin | 4616 (35.07) | 3368 (31.27) |
|
Other lipid lowering | 59 (0.45) | 32 (0.30) |
|
Blood pressure lowering | 5690 (43.23) | 4193 (38.93) |
|
Glucose lowering | 1065 (8.09) | 685 (6.36) |
|
Antithrombotic | 5863 (44.55) | 4329 (40.19) |
aLDL-c: low-density lipoprotein cholesterol.
Low-density lipoprotein cholesterol distributions stratified for patients with and without repeated measurements. A. Patients without repeated measurements. B. Patients with repeated measurements. Values on the x-axes represent mmol/L from the target.
In multivariable logistic regression analysis, more women were off target (odds ratio [OR] 1.48, 95% CI 1.40-1.56) compared to men (see
Logistic regression: factors associated with being off target at first measurement.
Characteristic | Odds ratio (95% CI)a | |
Age (per-year increase) | 0.99 (0.98-0.99) | |
Women | 1.48 (1.40-1.56) | |
Diabetes | 0.69 (0.55-0.65) | |
Hypertension | 0.87 (0.83-0.92) | |
Chronic kidney disease | 0.75 (0.54-1.04) | |
|
|
|
|
Statin use | 0.86 (0.80-0.93) |
|
Antithrombotic | 0.98 (0.91-1.05) |
Smoking | 1.29 (1.19-1.41) | |
Repeated measurements | 1.25 (1.19-1.32) |
aTotal number of patients was 23,932.
We extracted 51,383 repetitive measurements from 10,771 patients. Of these, 12,423 measurements (24.18%) were unrelated and, thus, excluded, leaving only one measurement for 2990 patients, which were also excluded. The number of measurements ranged from 2 to 40. After truncation of the measurements at the 75th percentile (number of measurements was 6), 25,438 LDL-c measurements in 7781 patients remained for the cluster analysis. State sequences, of which an example of 10 is shown in panel A from
State sequences of low-density lipoprotein cholesterol (LDL-c) categories. A. Example of the sequences from the first 10 patients in the dataset (10 seq.). B. State distributions (equal to prevalence of LDL-c categories) per measurement. C. Most common sequences. LDL-c values in the legend are in mmol/L. Cum % freq: cumulative percentage frequency; Freq: frequency.
The transition probabilities are shown in
Among these patients with related repeated measurements (N=10,771), 11,447 of 25,438 (45.00%) follow-up measurements remained on target or decreased to below the target threshold. We were able to assess factors associated with less favorable LDL-c change (ie, LDL-c that is stable but off target, decreased but not yet on target, or an increase in LDL-c) in a subset of 6871 measurements due to missing data on reported statin use (see
Transition probabilities for low-density lipoprotein cholesterol (LDL-c) categories between measurement pairs.
LDL-c category at |
LDL-c category at next measurement, transition probabilityb | |||||
|
On target | <0.5 mmol/L | 0.5-0.9 mmol/L | 1.0-1.4 mmol/L | 1.5-1.9 mmol/L | >2.0 mmol/L |
On target | 0.84 | 0.10 | 0.03 | 0.01 | 0.01 | 0.01 |
<0.5 mmol/L | 0.30 | 0.52 | 0.12 | 0.04 | 0.02 | 0.01 |
0.5-0.9 mmol/L | 0.23 | 0.19 | 0.43 | 0.09 | 0.03 | 0.02 |
1.0-1.4 mmol/L | 0.20 | 0.12 | 0.16 | 0.39 | 0.08 | 0.05 |
1.5-1.9 mmol/L | 0.19 | 0.13 | 0.11 | 0.12 | 0.36 | 0.10 |
>2.0 mmol/L | 0.15 | 0.13 | 0.09 | 0.10 | 0.09 | 0.43 |
aThe first measurement can be the first in a sequence as a whole or the first of a pair of measurements (eg, from the fourth to the fifth measurement).
bThe transition probability is the probability a patient will be in one of the LDL-c categories at next measurement given the last measurement, which is the first of the pair.
Logistic regression associations with deterioration of low-density lipoprotein cholesterol (LDL-c).
Characteristic | Odds ratio (95% CI)a | ||
Age (per-year increase) | 0.99 (0.99-1.00) | ||
Women | 1.44 (1.30-1.59) | ||
Diabetes | 0.72 (0.63-0.82) | ||
Hypertension | 0.93 (0.84-1.03) | ||
Smoking (current) | 1.00 (0.53-1.86) | ||
|
|
||
|
Same dose, same type | Reference | |
|
Same dose, different type | 0.81 (0.58-1.13) | |
|
Higher dose, same type | 1.82 (1.39-2.37) | |
|
Lower dose, same type | 1.31 (0.93-1.85) | |
|
Higher dose, different type | 1.47 (1.28-1.70) | |
|
Lower dose, different type | 0.92 (0.80-1.06) | |
Antithrombotic medication | 0.81 (0.73-0.89) | ||
Number of measurement | 0.98 (0.93-1.03) | ||
|
|
||
|
Short-term | Reference | |
|
Long-term | 0.97 (0.88-1.08) |
aTotal number of patients was 6871.
We evaluated the potential of routine clinical care data extracted from EHRs to obtain reliable information on LDL-c management in CVD patients referred to a tertiary care center. This approach may facilitate the implementation of a learning health care system, in which there is a constant cycle of data assembly, data analysis, interpretation, feedback, and change implementation. We showed that 51% of patients were not at their LDL-c target values at the time of referral. From a large proportion of patients, no follow-up LDL-c measurements (55%) were collected in our center. Patients with repeated measurements mostly showed no change in LDL-c level over time. The timing of drug prescription was difficult to determine from our data, limiting the interpretation of results regarding medication management.
Cardiovascular risk management, including LDL-c management, could substantially benefit from longitudinal evaluation of individual treatment trajectories. Cross-sectional studies, such as the EUROASPIRE IV, reported lower LDL-c target attainment compared to our findings [
Despite the compelling scientific evidence for the efficacy of LDL-c lowering in secondary prevention [
In our data, most patients remained in the same LDL-c category during every follow-up measurement. Possibly, attention for LDL-c management is limited in our tertiary care center, primarily focused on the complexity of disease and its comorbidity and, thus, LDL-c management might be more often delegated to the general practitioner. The large proportion of unique measurements (55%) and the finding that lower LDL-c target attainment was seen at baseline in patients with repeated measurements support this. Furthermore, treatment adherence due to polypharmacy—common in a tertiary population—might be challenging in our population [
Our study has several strengths. We used routine clinical care data, including time and individual trajectories, for the evaluation of LDL-c management. We selected patients with manifest CVD without restrictions to phenotype—with a 100% accuracy of defining manifest CVD—treated in all departments within our center, making our results generalizable to a large population. We expected some confounding by indication, with patients with a higher LDL-c being more likely to be followed up in our center, which was confirmed by the difference at the first measurements. Yet, for our evaluation, this does not influence the validity but merely shows good clinical practice: complex patients with high LDL-c values are followed up in our specialist tertiary care center.
We also encountered some challenges. Our study population was based on LDL-c measurements and was selected based on diagnosis and intervention codes, which are incomplete due to registration issues as well as registration in different centers. This likely did not influence our results in terms of directions and magnitude of the outcome measures, yet decreased the sample size of the study population. Future analysis could possibly take the patient as a starting point, first selecting all patients with CVD and then extracting LDL-c data from these patients. This would enable the reporting, also, of the number of patients in whom LDL-c was not measured. Furthermore, 55% of our patients were only measured once; from our data, we cannot determine whether this was due to insufficient management or a change in clinician that was responsible for the CVRM. Information on discontinuation of care within our center is unavailable; this calls for combining different data sources, including general practitioner and pharmacy data [
The EHR is a system primarily designed for registration of care. In clinical notes, clinicians register the clinical pathway of patients, including symptoms, measurements, and considerations of treatments. These considerations, in particular (ie, interpretation of data that leads to decisions), are difficult to capture within data extractions from the EHR. Harmonized clinical pathways with special attention to structured data collection are key for the availability and extractability of reliable data. Therefore, The Center for Circulatory Health of the UMC Utrecht initiated the Utrecht Cardiovascular Cohort (UCC) [
In conclusion, routine clinical care data can be used to obtain insights into clinical questions such as LDL-c target attainment and can be tailored into feedback from individual patients and clinicians. Our routine clinical care data, with high off-target prevalence, insufficient uptake of the guideline change, and little change in LDL-c over time, showed that improvement in guideline adherence is needed. Registrations of diagnosis, follow-up trajectory, and medication use need to be improved in order to enhance the usability of the EHR system for these types of questions.
Supplementary tables: Tables MA1-1 and MA1-2.
computerized decision support system
chronic kidney disease
cardiovascular disease
cardiovascular risk management
electronic health record
European Action on Secondary and Primary Prevention by Intervention to Reduce Events
low-density lipoprotein cholesterol
learning health care system
odds ratio
SUrvey of Risk Factor management
Utrecht Cardiovascular Cohort
University Medical Center
Utrecht Patient-Oriented Database
WWVS, IEH, SH, and Mark de Groot are members of the UPOD study group. The following are members of the UCC-CVRM Study group: FWA, Department of Cardiology; GJ de Borst, Department of Vascular Surgery; MLB (chair), Julius Center for Health Sciences and Primary Care; S Dieleman, Division of Vital Functions (anesthesiology and intensive care); MH Emmelot, Department of Geriatrics; PA de Jong, Department of Radiology; ATL, Department of Obstetrics and Gynecology; IEH, Laboratory of Clinical Chemistry and Hematology; NP van der Kaaij, Department of Cardiothoracic Surgery; YM Ruigrok, Department of Neurology; MC Verhaar, Department of Nephrology and Hypertension; and FLJ Visseren, Department of Vascular Medicine, UMC Utrecht and Utrecht University.
This project was financially supported in part by Sanofi. DK is currently a full-time employee of Sanofi-Aventis. The remaining authors declare no conflict of interest.