Targeted identification of adverse events in coronary artery disease patients based on patient-reported outcomes
Abstract
Aim: Can focusing the adverse events search to patients with poor patient-reported outcome help in targeting adverse event detection? Patients & methods: Coronary artery revascularization patients of the Kuopio University Hospital from June 2012 to August 2014 categorized into those with clinically significant improvement (15D score change ≥0.015, n = 81) or deterioration (change ≥-0.015, n = 64) in post-intervention health-related quality of life. Results: Major complications (27 vs 9%, p = 0.004) or post-intervention infections (16 vs 5%, p = 0.031) were more common among those with deteriorated score. They also tended to have more cardiovascular (19 vs 9%, p = 0.071) and minor complications (16 vs 7%, p = 0.118). Conclusion: Patient-reported outcomes may potentially help in targeting the adverse events search so that a larger number of adverse events can be identified for efficient learning from them.
To be able to continuously improve healthcare, identification of adverse events and learning from them is important. Voluntary reporting systems have gained popularity during recent years but underestimate the real number of adverse events as only a small percentage of them are usually reported [1,2]. The same seems to apply, at least in Finland, to the use of International Classification of Diseases codes generated for reporting of adverse events as they in practice appear to be rarely recorded.
Reviewing patient records is currently considered the best way to detect adverse events, but it is tedious, time consuming and often not very productive. To advance the identification of adverse events from patient records, the Institute for Healthcare Improvement [3] has developed the Global Trigger Tool which is currently considered as the golden standard for detecting adverse events in patient safety research [4–7]. Global Trigger Tool aims to narrow down, by identifying certain keywords associated with safety incidents, the number of full patient records that needs to be reviewed [8].
An alternative or complementary approach could be to focus the searching of adverse events to patients with poor patient-reported treatment outcomes (PROs). Adverse events usually affect the subjective state of health negatively and, consequently, are often reflected in poor PROs. As utility of treatment is currently often monitored by using PROs, data collected for this purpose could possibly also be used for the identification of patients that are more likely than average to have suffered an adverse event.
Our hospital collects routine PRO data in the form of health-related quality of life (HRQoL) measurements which makes it also possible to easily identify patients in whom the PRO does not improve following treatment. We tested in a pilot study on coronary artery disease patients whether there is an association between poor PROs and adverse events and/or comorbidities in the hope that such an approach would facilitate focusing the search for adverse events on the right patients.
Patients & methods
Study design & setting
Our hospital has collected HRQoL data with the 15D instrument [9–11] as part of its routine practice since autumn 2011. Currently such data is collected in 16 distinct patient groups. The participants of this study were cardiac patients admitted for coronary revascularization therapy (coronary artery bypass grafting or percutaneous coronary intervention). They were, from June 2012 to August 2014, asked, as part of routine clinical practice, to fill in the 15D questionnaire at baseline and 12 months after the revascularization procedure.
The 15D instrument is a generic, self-administered HRQoL questionnaire. It consists of 15 dimensions (mobility, vision, hearing, breathing, sleeping, eating, speech, excretion, usual activities, mental function, discomfort and symptoms, depression, distress, vitality and sexual activity) with five ordinal levels. The 15D instrument produces single index scores which range from 0 to 1. It can generate over 30 billion different health states. The valuation system of the 15D used in this study is based on a set of population-based preferences [9–11]. The minimal important difference (MID) in the 15D score has been reported to be ± 0.015 [12].
Patients having returned both the baseline and the 12-month questionnaires were categorized, based on the observed changes in their HRQoL scores after the treatment, to those with a clinically significant improvement (change in the 15D score ≥ 0.015), or a clinically significant deterioration (change in the 15D score ≥ -0.015) [12]. Patient records of both groups were then reviewed for possible comorbidities, both before and after the intervention and adverse events during the follow-up by two experienced investigators who were unaware of the treatment outcome.
Adverse events were categorized into clinically significant (e.g., those requiring reoperation, extended hospital stay or rehospitalization, neurologic complications with devastating consequences such as stroke, etc.), or minor (e.g., uncomplicated wound infections) adverse events (Table 1). In case a patient had encountered both a major and a minor complication, he/she was classified only into the major complication category. All reported comorbidities were noted and categorized into those deemed clinically significant from a HRQoL point of view (e.g., neurologic comorbidities with a physical handicap or severe heart or lung conditions restricting even light physical exercise, etc.) or nonsignificant (hypertension, uncomplicated diabetes etc.; see Table 1).
| Adverse event type | Clinically significant | Minor/clinically nonsignificant |
|---|---|---|
| Cardiovascular complication | Re-operation required | Postoperative hematoma in groin |
| Extended hospital stay or | Arrhythmia | |
| rehospitalization | Postoperative pneumothorax | |
| Neurologic complication (stroke) | Postoperative atrial fibrillation | |
| Stent thrombosis during follow-up | – | |
| Myocardial infarction | – | |
| Embolus | – | |
| Infection | Sternum wound infection | Operation wound excretion |
| Sternum wound revision | Other wound infection | |
| Sternum wound dehiscence | – | |
| Sepsis | – | |
| Urosepsis | – | |
| Other complication | Failed revascularization | Postoperative ileus |
| Postoperative mental confusion | Ileus | |
| Hip luxation | Rash (Plavix) | |
| Peroneal paresis | – | |
| Comorbidities in the study population | Neurologic comorbidities with a physical handicap | Hypertension |
| Severe heart condition/heart failure | Uncomplicated diabetes | |
| Severe lung condition | Asthma | |
| Claudication | Cancer having been treated earlier | |
| Spinal stenosis | Tinnitus | |
| Chronic obstructive pulmonary disease | Benign musculoskeletal disorders | |
| Spinal disc herniation during follow-up | Benign prostatic hyperplasia | |
| New cancer during follow-up | Eye pain | |
| Continuous pain | Collapse | |
| Ulcerative colitis | Rheumatic disorder | |
| Asbestosis | Pneumonia | |
| Stroke or other cerebrovascular incident | Parkinson's disease | |
| Recurrent pneumonia | Metabolic syndrome | |
| Complicated diabetes | Need for pacemaker during follow-up | |
| Symptomatic lower limb atherosclerosis | Chronic atrial fibrillation |
As our university hospital serves as a tertiary care center and is responsible for invasive cardiac interventions of also some hospital districts outside of its immediate catchment area, post-intervention treatment of some of the cardiac patients took place in other hospitals. To ensure that we had access to patient records covering the whole secondary care treatment path of the patients, only patients whose entire secondary care treatment took place in this university hospital, were included in the analysis.
Statistical analyses
Descriptive statistics for demographic data are presented using frequencies and percentages. Results are presented as means with 95% confidence intervals. Differences between the groups of interests were tested by ordinary least square regression analysis. Differences between groups with a positive and a negative PRO were tested in a univariate logistic regression model using age, sex, operation type, number of all comorbidities and the absence or presence of significant comorbidities as independent variables. Multivariable logistic regression was performed using the same variables as above, except for the number of comorbidities because of its collinearity with the presence or absence of significant comorbidities. The observed HRQoL score changes were normally distributed and associations between various characteristics and HRQoL changes were tested by linear regression. All statistical analyses were conducted by STATA 12.0 (Stata Corp. LP, TX, USA) or SPSS 19 (IBM SPSS statistics for Windows, NY, USA).
Results
During the study period, the mean 15D score change for all 378 revascularized patients having answered both the baseline and the 12-month 15D questionnaire was 0.016 (0.022 for women and 0.014 for men, p = 0.442). Age was inversely associated with the 15D score gain (Table 2).
| Population | n (%) | Mean 15D score change (95% CI) | Least square difference (95% CI) | p-value† |
|---|---|---|---|---|
| Men | 303 (80.2) | 0.014 (0.005–0.023) | (Reference) | |
| Women | 75 (19.8) | 0.022 (0.005–0.039) | 0.008 (-0.013–0.029) | 0.442 |
| CABG | 209 (55.3) | 0.029 (0.019–0.040) | (Reference) | |
| PCI | 169 (44.7) | -0.001 (-0.014–0.011) | -0.031 (-0.048 to -0.015) | <0.001 |
| Age under 60 | 89 (23.5) | 0.027 (0.010–0.044) | (Reference) | |
| Age 60–74.9 | 203 (53.7) | 0.018 (0.009–0.029) | -0.009 (-0.029–0.012) | 0.412 |
| Age 75 and over | 86 (22.8) | -0.004 (-0.025–0.017) | -0.031 (-0.056 to -0.007) | 0.011 |
| No MID | 73 (19.3) | 0.0004 (-0.001–0.002) | (Reference) | |
| Positive MID | 185 (48.9) | 0.080 (0.073–0.088) | 0.080 (0.068–0.922) | <0.001 |
| Negative MID | 120 (31.8) | -0.075 (-0.083 to -0.066) | -0.075 (-0.088 to -0.062) | <0.001 |
†Statistical significance of difference estimated by ordinary least squares regression.
CABG: Coronary artery bypass grafting; MID: Minimal important difference; PCI: Percutaneous coronary intervention.
Altogether 145 patients with a clinically significant 15D score change lived in the immediate catchment area of the hospital. 81 (55.9%) of them had a positive MID and 64 (44.1%) a negative MID. The group with a minimal clinically significant negative 15D score change had, on average, more adverse events or comorbidities than the group with clinically significantly improved 15D score. The patients with a clinically significant positive 15D score change were on average 3 years younger than those with a negative change (65.4 vs 68.3 years, p = 0.071; Table 3). The proportion of women was somewhat higher in the group with a positive change than in the group with a negative change, but the difference did not reach statistical significance (28.4 vs 18.8%, respectively, p = 0.180).
| Groups | n (%) | MID, n (%) | Unadjusted | Adjusted† | |||||
|---|---|---|---|---|---|---|---|---|---|
| n = 145 (100.0) | Positive n = 81 (55.9) | Negative n = 64 (44.1) | OR | 95% CI | p-value‡ | OR | 95% CI | p-value‡ | |
| Mean age (range) | 66.7 (47–90) | 65.4 (47–89) | 68.3 (49–90) | 0.967 | 0.932–1.003 | 0.071 | 0.966 | 0.928–1.005 | 0.083 |
| Men | 110 (75.9) | 58 (71.6) | 53 (81.3) | 1.000 | (Reference) | 1.000 | (Reference) | ||
| Women | 35 (24.1) | 23 (28.4) | 12 (18.7) | 1.718 | 0.778–3.794 | 0.180 | 2.751 | 1.112–6.805 | 0.028 |
| Comorbidities | |||||||||
| 0 | 10 (6.9) | 7 (8.6) | 3 (4.7) | 1.000 | (Reference) | Omitted due to collinearity§ | |||
| 1 | 29 (20.0) | 19 (23.5) | 10 (15.6) | 0.814 | 0.172–3.853 | 0.796 | |||
| 2 | 44 (30.3) | 30 (37.0) | 14 (21.9) | 0.918 | 0.206–4.091 | 0.911 | |||
| 3 | 24 (16.6) | 11 (13.6) | 13 (20.3) | 0.363 | 0.075–1.748 | 0.206 | |||
| 4 | 22 (15.2) | 7 (8.6) | 15 (23.4) | 0.200 | 0.039–1.014 | 0.052 | |||
| ≥5 | 16 (11.0) | 7 (8.6) | 9 (14.1) | 0.333 | 0.062–1.779 | 0.199 | |||
| Significant comorbidities | 46 (31.7) | 11 (18.5) | 31 (48.4) | 0.242 | 0.115–0.509 | <0.001 | 0.215 | 0.097–0.476 | <0.001 |
| Operation type | |||||||||
| PCI | 96 (66.2) | 51 (63.0) | 45 (70.3) | 1.000 | (Reference) | ||||
| CABG | 49 (33.8) | 30 (37.0) | 19 (29.7) | 1.393 | 0.691–2.807 | 0.352 | 1.332 | 0.168–2.868 | 0.464 |
†Adjusted logistic regression age, sex and significant comorbidities.
‡Statistical significance of difference estimated by unadjusted and adjusted logistic regressions.
§Significant comorbidities and number of comorbidities. Spearman correlation was 0.507. Because of collinearity, a number of comorbidities were excluded to adjusted logistic regression model.
CABG: Coronary artery bypass grafting; PCI: Percutaneous coronary intervention.
Of the patients with a clinically significant negative change in their 15D score, 26.6% had encountered a major complication compared with 8.6% of those with a positive treatment result (p = 0.004; see Figure 1A). The same was true for post-intervention infections (15.6 vs 4.9%, respectively, p = 0.031; see Figure 1B). Patients with a poor PRO also tended to have more often minor complications (15.6 vs 7.4%, respectively, p = 0.118; see Figure 1A) and cardiovascular complications (18.8 vs 8.6%, respectively, p = 0.071; see Figure 1B) but the differences were not statistically significant. There were no significant differences in the number of comorbidities between the groups. However, comorbidities which were deemed clinically significant from a HRQoL perspective were significantly more frequent in the group with poor PRO (18.5 vs 48.4%, p <0.001; see Table 3). The association between adverse events and clinically significant 15D score change, nevertheless, remained after adjusting for comorbidities (Table 3).
Figure 1. Percentages of patients with complications.
(A) Major and minor complications and (B) cardiovascular complications, infections and other complications among those who experienced a clinically significant improvement (positive MID) or deterioration in health-related quality of life (negative MID) after coronary artery revascularization procedure. Significance of differences between positive and negative MID groups were calculated by ordinary least square univariate regressions.
MID: Minimally important difference.
The mean 15D change was negative both in patients with minor (-0.013, 95% CI: -0.067–0.419) or major adverse events (-0.029, 95% CI: -0.061–0.004; see Table 4). Furthermore, in patients with cardiovascular or infectious complications, the mean 15D score change was always negative although the difference, compared with patients without those complications, reached statistical significance only in the case of major complications and complications that were deemed to be unrelated to the revascularization therapy. Patients with significant comorbidities also showed a statistically significant deterioration in their mean 15D score (-0.025, 95% CI: -0.173–0.182) compared with the rest of the patients (Table 4).
| Groups | n (%) | Mean 15D change (95% CI) | Least square difference (95% CI) | p-value† |
|---|---|---|---|---|
| Women | 35 (24.1) | 0.031 (0.002–0.061) | (Reference) | – |
| Men | 110 (75.9) | 0.006 (-0.011–0.023) | 0.025 (-0.009–0.059) | 0.154 |
| Operation type: | ||||
| – PCI | 96 (66.2) | 0.003 (-0.015–0.021) | (Reference) | – |
| – CABG | 49 (33.8) | 0.030 (0.004–0.056) | 0.027 (-0.004–0.058) | 0.089 |
| Age under 60 | 45 (31.0) | 0.024 (-0.003–0.051) | (Reference) | – |
| Age 60–74.9 | 63 (43.5) | 0.020 (-0.001–0.182) | -0.003 (-0.038–0.031) | 0.842 |
| Age 75 and over | 37 (25.5) | -0.016 (-0.047–0.016) | -0.039 (-0.079–0.001) | 0.048 |
| No minor complications | 129 (89.0) | 0.015 (0.001–0.030) | (Reference) | – |
| Minor complication | 16 (11.0) | -0.013 (-0.067–0.419) | -0.028 (-0.075–0.019) | 0.240 |
| No major complications | 121 (83.4) | 0.020 (0.004–0.037) | (reference) | – |
| Major complication | 24 (16.6) | -0.029 (-0.061–0.004) | -0.049 (-0.088 to -0.010) | 0.014 |
| No infection | 131 (90.3) | 0.016 (0.001–0.032) | (Reference) | – |
| Infection | 14 (9.7) | -0.028 (-0.069–0.012) | -0.045 (-0.094–0.005) | 0.076 |
| No cardiovascular complications | 126 (86.9) | 0.015 (-0.001–0.031) | (Reference) | – |
| Cardiovascular complication | 19 (13.2) | -0.007 (-0.049–0.036) | -0.022 (-0.066–0.022) | 0.327 |
| No other complications | 135 (93.1) | 0.016 (0.001-0.032) | (Reference) | – |
| Other complication | 10 (6.9) | -0.042 (-0.096–0.012) | -0.059 (-0.116 to -0.001) | 0.047 |
| Number of comorbidities | ||||
| 0 | 10 (6.9) | 0.026 (-0.103–0.158) | (Reference) | – |
| 1 | 29 (20.0) | 0.022 (-0.199–0.145) | -0.004 (-0.069–0.061) | 0.906 |
| 2 | 44 (30.3) | 0.029 (-0.151–0.175) | 0.003 (-0.059–0.065) | 0.916 |
| 3 | 24 (16.6) | 0.016 (-0.138–0.230) | -0.010 (-0.076–0.057) | 0.774 |
| 4 | 22 (15.2) | -0.024 (-0.153–0.165) | -0.050 (-0.117–0.017) | 0.143 |
| ≥5 | 16 (11.0) | -0.013 (-0.173–0.182) | -0.039 (-0.110–0.032) | 0.277 |
| No significant comorbidities | 99 (68.3) | 0.030 (0.014–0.046) | (Reference) | – |
| Significant comorbidities | 46 (31.7) | -0.025 (-0.173–0.182) | -0.055 (-0.085 to -0.024) | <0.001 |
†Ordinary least square univariate regression models produced for significance test of differences.
CABG: Coronary artery bypass grafting; PCI: Percutaneous coronary intervention.
Discussion
According to the results of this pilot study on coronary artery disease patients, there is a clear negative association between adverse events and HRQoL gain in patients undergoing coronary revascularization therapy. This reflects the 15D instrument's sensitivity to detect factors which affect the overall subjective state of health negatively. Partly the poor PRO can be explained by the higher number of comorbidities in the group with a negative outcome. However, the statistically significant association between the adverse events and negative clinically significant 15D score change remained after adjusting for comorbidities.
Using the PRO of treatment as measured by the change in HRQoL could, according to our results, be a complementary approach for targeting the search for adverse events to those patients most likely to have encountered them. Identifying and measuring harm have been seen as core patient safety goals and the identification of vulnerabilities is necessary to learn from mistakes and to be able to take corrective action [13].
A limitation of the study is the fact that our preliminary findings are based on a small number of patients and only on coronary artery disease patients. The findings should thus be considered as tentative. As the 15D seems to react also to the occurrence of major infections, it is likely to help detect a wide range of unreported adverse events and also other than those directly related to the intervention. As HRQoL measurements are currently a fundamental part of routine comparative effectiveness analyses in our university hospital, they could, in addition to monitoring the effectiveness and cost–effectiveness of treatment, also be used to help focus the detection of adverse events to the right patients. Most likely, our approach could also utilize data obtained with other HRQoL instruments, but currently we have no results to substantiate such a claim. Furthermore, the generalizability of our results to other centers or settings needs to be established in future studies. The more adverse events we are able to identify, the more we can learn from them and thus develop safer and more effective hospital practices. Furthermore, analysis of patients with a positive or a negative HRQoL change can hopefully in the future improve the selection of patients most likely to benefit from revascularization therapy and, on the other hand, the identification of those that are prone to develop complications and comorbidities and, thus unlikely to gain from revascularization.
A strength of this pilot study is the fact that the findings are based on real-life data routinely collected in the hospital, not on highly selected patients usually seen in randomized controlled trials. Consequently, our results can probably be generalized also to other settings but further studies are needed to see whether similar results can also be obtained in other patient groups.
Conclusion
Poor effectiveness of treatment, as judged by a PRO, is often associated with adverse events. Using routinely collected PRO data may help target the search for adverse events to the right patients, and may potentially reveal a larger number of adverse events than currently used approaches. This enables more targeted improvement of healthcare processes. In the future, wider use of HRQoL measurements for the detection adverse events could be a new approach for harm reduction in healthcare.
Identification of adverse events and learning from them, is essential for improving the quality of healthcare.
Current approaches have drawbacks as the voluntary reporting systems underestimate the real number of adverse events and reviewing patient records is tedious, time consuming and often not very productive.
Adverse events usually affect the subjective state of health negatively and are consequently often reflected in poor patient-reported health outcomes.
Major complications (27 vs 9%, p = 0.004) or post-intervention infections (16 vs 5%, p = 0.031) were more common among those coronary artery revascularization patients with deteriorated score than patients with improved score.
Patients with a deteriorated 15D score also tended to have more cardiovascular (19 vs 9%, p = 0.071) and minor complications (16 vs 7%, p = 0.118).
Focussing the searching of adverse events to patients with poor patient-reported treatment outcome may help target the search for adverse events to the right patients. This enables more efficient learning from adverse events and consequent improvement of healthcare processes.
Analysis of patients with a positive or a negative change in health-related quality of life can also improve the selection of patients most likely to benefit from revascularization therapy and, on the other hand, the identification of those that are prone to develop complications and comorbidities and, thus unlikely to gain from revascularization.
Comparative effectiveness data can play a major role in improving patient safety.
Financial & competing interests disclosure
J Martikainen is a partner of ESiOR Oy, which carries out health economic and outcome research studies for pharmaceutical, medical device and food companies. The authors have no other relevant affiliations or financial involvement with any organization or entity with a financial interest in or financial conflict with the subject matter or materials discussed in the manuscript apart from those disclosed.
No writing assistance was utilized in the production of this manuscript.
Ethical conduct of research
Standard protocol approvals, registrations and patient consents. The study was approved by the ethics committee of the hospital. All data used in the study were obtained by routine data collection of the hospital and the patients were not approached in any way because of the study. As all information was collected as a part of routine treatment, informed consent was not required by the ethics committee. The study complies with the Declaration of Helsinki.
References
Papers of special note have been highlighted as: • of interest; •• of considerable interest
1.
Sari AB, Sheldon TA, Cracknell A, Turnbull A. Sensitivity of routine system for reporting patient safety incidents in an NHS hospital: retrospective patient case note review. BMJ 334(7584), 79–81 (2007).
2.
Rafter N, Hickey A, Condell S et al. Adverse events in healthcare: learning from mistakes. QJM 108(4), 273–277 (2015).
•• Stresses the importance of learning from mistakes but also the lack of consensus concerning the detection of adverse events.
3.
Institute for Healthcare Improvement. www.ihi.org.
4.
Naessens JM, O'Byrne TJ, Johnson MG, Vansuch MB, McGlone CM, Huddleston JM. Measuring hospital adverse events: assessing inter rater reliability and trigger performance of the Global Trigger Tool. Int. J. Qual. Health Care 22(4), 266–274 (2010).
• Brings forth the Global Trigger Tool for assessing adverse events and provides information on its reliability.
5.
Härkänen M, Turunen H, Vehviläinen-Julkunen K. Differences between methods of detecting medication errors: a secondary analysis of medication administration errors using incident reports, the global trigger tool method, and observations. J. Patient Saf. (2016) (Epub ahead of print).
6.
Rutberg H, Risberg MB, Sjödahl R, Nordqvist P, Valter L, Nilsson L. Characterisations of adverse events detected in a university hospital: a 4-year study using the Global Trigger Tool method. BMJ Open 4, e004879 (2014).
• Discusses the limitations of voluntary reporting systems in reporting adverse events in hospitals.
7.
Deilkås ET, Bukholm G, Lindstrøm J, Haugen M. Monitoring adverse events in Norwegian hospitals from 2010 to 2013. BMJ Open 5, e008576 (2015).
• Discusses the need to improve adverse event detection in Norwegian hospitals.
8.
Thomas EJ, Petersen LA. Measuring errors and adverse events in healthcare. J. Gen. Intern. Med. 18(1), 61–67 (2003).
•• Clarifies the strengths and weaknesses of assessing adverse events using different sources.
9.
Sintonen H. The 15D-measure of health-related quality of life. I. Reliability, validity and sensitivity of its health state descriptive system. National Centre for Health Programme Evaluation, Working Paper 41, Melbourne, Australia (1994).
10.
Sintonen H. The 15D-measure of health-related quality of life. II. Feasibility, reliability and validity of its valuation system. National Centre for Health Programme Evaluation, Working Paper 42, Melbourne, Australia (1995).
11.
Sintonen H. The 15D instrument of health-related quality of life: properties and applications. Ann. Med. 33(5), 328–336 (2001).
12.
Alanne S, Roine RP, Räsänen P, Vainiola T, Sintonen H. Estimating the minimum important change in the 15D scores. Qual. Life Res. 24(3), 599–606 (2015).
•• Reports the minimal important difference value for the 15D health-related quality of life instrument.
13.
Vincent C, Burnett S, Carthey J. The Measurement and Monitoring of Safety. Health Foundation, London, UK (2013).
Information & Authors
Information
Published In
Copyright
© 2017 Future Medicine Ltd.
History
Received: 7 December 2016
Accepted: 25 July 2017
Published online: 17 October 2017
Keywords:
Topics
Authors
Metrics & Citations
Metrics
Article Usage
Article usage data only available from February 2023. Historical article usage data, showing the number of article downloads, is available upon request.
Citations
How to Cite
Targeted identification of adverse events in coronary artery disease patients based on patient-reported outcomes. (2017) Journal of Comparative Effectiveness Research. DOI: 10.2217/cer-2016-0091
Export citation
Select the citation format you wish to export for this article or chapter.