Category: Cardiology

The Machine Can Learn

A couple weeks ago I covered computerized diagnosis via symptom checkers, noting their imperfect accuracy – and grossly underperforming crowd-sourced physician knowledge. However, one area that continues to progress is the use of machine learning for outcomes prediction.

This paper describes advances in the use of “big data” for prediction of 30-day and 180-day readmissions for heart failure. The authors used an existing data set from the Telemonitoring to Improve Heart Failure Outcomes trial as substrate, and then applied several unsupervised statistical models to the data with varying inputs.

There were 236 variables available in the data set for use in prediction, weighted and cleaned to account for missing data. Compared with the C statistic from logistic regression as their baseline comparator, the winner was pretty clearly Random Forests. With a baseline 30-day readmission rate of 17.1% and 180-day readmission of 48.9%, the C statistic for the logistic regression model predicting 30-day readmission was 0.533 – basically no predictive skill. The Random Forest model, however, achieved a C statistic of 0.628 by training on the 180-day data set.

So, it’s reasonable to suggest there are complex and heterogenous data for which machine learning methods are superior to traditional models. These are, unfortunately, pretty terrible C statistics, and almost certainly of very limited use for informing clinical care. As with most decision-support algorithms, I would be curious also to see a comparison with a hypothetical C statistic for clinician gestalt. However, for some clinical problems with a wide variety of influential factors, these sorts of models will likely become increasingly prevalent.

“Analysis of Machine Learning Techniques for Heart Failure Readmissions”
http://circoutcomes.ahajournals.org/content/early/2016/11/08/CIRCOUTCOMES.116.003039

All Glory to the Triple-Rule-Out

The conclusions of this study are either ludicrous, or rather significant; the authors are either daft, or prescient. It depends fundamentally on your position regarding the utility of CT coronary angiograms.

This article describes a retrospective review of all the “Triple-Rule-Out” angiograms performed at a single center, Thomas Jefferson University Hospital, between 2006 and 2015. There were no specific circumstances under which the TRO were performed, but, grossly, the intended population were those who were otherwise being evaluated for an acute coronary syndrome but “was suspected of having additional noncoronary causes of chest pain”.

This “ACS-but-maybe-not” cohort totaled 1,192 patients over their 10 year study period. There were 970 (81.4%) with normal coronary arteries and no significant alternative diagnosis identified. The remaining, apparently to these authors, had “either a coronary or noncoronary diagnosis that could explain their presentation”, including 139 (11.7%) with moderate or severe coronary artery disease. In a mostly low-risk, troponin-negative population, it may be a stretch to attribute their symptoms to the coronary artery disease – but I digress.

The non-coronary diagnoses, the 106 (8.6%) with other findings, range from “important” to “not at all”. There were, at least, a handful of aortic dissections and pulmonary emboli picked up – though we can debate the likelihood of true positives based on pretest odds. However, these authors also credit the TRO with a range of sporadic findings as diverse as endocarditis, to diastasis of the sternum, and 24 cases of “aortic aneurysm” which were deemed important mostly because there were no priors for comparison.

The authors finally then promote TRO scans based on these noncoronary findings – stating that, if a traditional CTCA were performed, many of these diagnosis would likely be missed. Thus, the paradox. If you are already descending the circles of hell, and are using CTCA in the Emergency Department – then, yes, it is reasonable to suggest the TRO is a valid extension of the CTCA. Then again, if CTCA in the acute setting is already outside the scope of practice, and TRO is an abomination – carry on as if this study never existed.

“Diagnostic Yield of Triple-Rule-Out CT in an Emergency Setting”
http://www.ncbi.nlm.nih.gov/pubmed/27186867

Beware Good Times; Beware!

I imagine some of you are active travelers, eating on the go. Perhaps you have business interests and frequently dine out of the home. Sadly, your rock and roll lifestyle may yet be the end of you.

This is an interesting analysis of the “Progression of Early Subclinical Atherosclerosis” study, breaking down participants into dietary spectra. Based on cross-sectional analysis of the reported eating habits of participants, there were three distinct dietary patterns cohorted together: Mediterranean, Western, and and a “social-business” eating pattern. Approximately 40% were in each the Mediterranean and Western cohorts, with the remainder in “social-business”. The general attributes of each diet are shown in the figure below:

eating patterns image

The finding of note – after all participants underwent atherosclerotic screening tests – is that the “social-business” pattern of consumption is, quite obviously, the highest risk for disease progression. Whereas the Mediterranean cohort had only 36% with generalized or intermediate atherosclerosis, the “social-business” cohort totaled an impressive 68%. The Western diet was slightly above the Mediterranean, at 41%.

Stay healthy, my friends.

“Association Between a Social-Business Eating Pattern and Early Asymptomatic Atherosclerosis”

http://content.onlinejacc.org/article.aspx?articleID=2544530

The High-Sensitivity Troponin Ennui

They’re coming. It’s inevitable. They have yet to be approved in the the United States, but every year the news is the same: they’re coming.

High-sensitivity troponins have been both lauded and mocked from various perspectives. The literature is replete with examples of expedited rule-outs in the Emergency Department owing to their improved lower limit of detection for myocardial injury. However, every study touting the benefits of improved sensitivity has begrudgingly or worse acknowledged the correspondingly diminished specificity.

This, then, is a randomized trial of reporting either a conventional troponin assay result or a high-sensitivity troponin assay result, with a multitude of patient-oriented short- and long-term outcomes measured. The specific assays used here were either a c-TnT with a threshold of detection of 30 ng/L, or a hs-TnT with a threshold of detection of 3 ng/L. Clinicians caring for patients were randomized to making care decisions based on one, without knowledge of the other.

For all the various propaganda for and against high-sensitivity troponins, this trial is highly anticlimactic. There were, essentially, no changes in physician behavior resulting from the additional information provided by the more sensitive assay. No fewer patients were admitted, similar numbers of ultimate downstream tests occurred, and there were no reliable differences in long-term cardiac or combined endpoint outcomes.

The only outcome of note is probably consistent with what we already knew: any circulating troponin portends worse outcomes. This may be most helpful in directing the long-term medical management of those whose troponin levels were previously undetectable with a conventional assay; these patients clearly do not have the same virtually-zero risk as a patient with undetectable troponin levels. Indeed, troponin levels alone were a better predictor of long terms outcomes than the Heart Foundation Risk Stratification, as well.

I’ll let Judd Hollander sum it up in his most concise – with a link to much more verbose – terms:

“Randomized Comparison of High-Sensitivity Troponin Reporting in Undifferentiated Chest Pain Assessment”
http://circoutcomes.ahajournals.org/content/early/2016/08/09/CIRCOUTCOMES.115.002488.abstract

Next Up in Syncope Prediction

The Great White North is the land of clinical decision instruments.  Canadian Head, Canadian C-Spine, Ottawa Ankle, Ottawa SAH, the list goes on – and now, from the same esteemed group: the Canadian Syncope Risk Score.

The vast majority of patients with syncope have unrevealing initial and, if admitted, in-house evaluation.  That said, any physiologic interruptions in the ability to perfuse the brain portend a poor prognosis greater than the general background radiation.  These authors performed an observational study over the course of four years to prospectively derive a decision instrument to support risk-stratification for syncope.

There were 4,030 patients enrolled and eligible for analysis based on 30-day follow-up, and 147 of these suffered a “serious adverse event”.  They identified 43 candidate predictors for prospective collection, and ultimately this resulted in a multivariate logistic regression predictive model with 9 elements.  Scores range from -3, with a 0.4% estimated risk for SAE, to 11, with an 83.6% estimated risk for SAE.  Useable confidence intervals, however, were mostly scores <5.

There are a few things I would quibble with regarding this study.  The “serious adverse event” definition is rather broad, and includes 30-day events for which the underlying pathology was not present or necessarily preventable at the initial visit.  For example, a patient with a subsequent encounter for a GI bleed or a case of appendicitis fit their criteria of SAE.  This would diminish the instrument’s apparent sensitivity without materially improving its clinical relevance.  Then, there is the oddity of incorporating the final ED diagnosis into the scoring system – where a provisional diagnosis of “vasovagal syncope” is -2, and a diagnosis of “cardiac syncope” is +2.  The authors explicitly defend its inclusion and the methods behind it – but I feel its subjectivity coupled with widespread practice variation will impair this rule’s generalizability and external validation.

Finally, the last issue with these sorts of “rules”: “high risk” is frequently conflated to mean “admit to hospital”.  In many situations close to the end-of-life, the protective effect of hospitalization and medical intervention vanishes – and may have little or no value.  This sort of stratification should be applied within the appropriate medical and social context, rather than simply triggering admission.

“Development of the Canadian Syncope Risk Score to predict serious adverse events after emergency department assessment of syncope”
http://www.ncbi.nlm.nih.gov/pubmed/27378464

High Blood Pressure is Not a Crime

And you don’t need to be sent to “time out” – i.e., referred to the Emergency Department – solely because of it.

This is a retrospective, single-center report regarding the incidence of adverse events in patients found to have “hypertensive urgency” in the outpatient setting.  This was defined formally as any systolic blood pressure measurement ≥180 mmHg or diastolic measurement ≥110 mmHg.  Their question of interest was, specifically, whether patients referred to the ED received clinically-important diagnosis (“major adverse cardiovascular events”), with a secondary interest in whether their blood pressure was under better control at future outpatient visits.

Over their five-year study period, there were 59,535 patient encounters meeting their criteria for “hypertensive urgency”.  Astoundingly, only 426 were referred to the Emergency Department.  Of those referred to the ED, 2 (0.5%) received a MACE diagnosis within 7 days, compared with 61 (0.1%) of the remaining 58,109.  By 6 months, MACE had equalized between the two populations – now 4 (0.9%) in the ED referral cohort compared with 492 (0.8%) in those sent home.  Hospital admission, obviously, was higher in those referred to the ED, but apparently conferred a small difference in blood pressure control in follow-up.

The authors go on to perform a propensity-matched comparison of the ED referrals to the sent home cohort, but this is largely uninsightful.  The more interesting observation is simply that these patients largely do quite well – and any adverse events probably happen at actuarial levels rather than having any specific relationship to the index event.

I appreciate how few patients were ultimately referred to the Emergency Department in this study; fewer than 1% is an inoffensive number.  That said, zero percent would be better.

“Characteristics and Outcomes of Patients Presenting With Hypertensive Urgency in the Office Setting”
http://archinte.jamanetwork.com/article.aspx?articleid=2527389

Zeno’s Zero-Hour Rule-Outs

This is TRAPID-AMI, a prospective Roche hs-TnT (99th percentile 14ng/L) study for patients presenting within 6 hours of peak chest pain symptoms.  Samples were drawn on arrival, then serially over the next two hours, and a final sample drawn 4-14h after presentation.  The primary outcome was acute MI, as adjudicated by a panel of cardiologists blinded to the hs-TnT result but aware of a Siemens conventional troponin I (99th percentile 40ng/L) result.  Interestingly, they powered their study specifically for a negative predictive value, rather than a specific sensitivity result – certainly an easier enrollment target, but also potentially more difficult to generalize.

All told, they included 1,282 patients in the study and there were 213 patients ultimately diagnosed with AMI on their initial presentation.  In 30-day follow-up, 11 patients could not be contacted, and an additional 18 patients not originally diagnosed with AMI met their combined MACE endpoint, 15 of which were revascularlization.  Of these 213 patients with initial diagnosis of AMI, there were four false-negatives with respect to the troponin assays being below the limit of detection.  For their primary outcome, then, sensitivity was 98.1% (95.3-99.5) with specificity 52.0% (49.0-55.0).  The authors go on to present several different strategies based on various cut-offs and various EKG findings, the effect of which are various trade-offs between number of patients eligible for a zero-hour rule-out, the sensitivity, and the specificity.  Negative predictive values, of course, generally ran >99%, owing the the relative infrequency of AMI diagnoses.

This study, like most before it, provides little additional insight.  Sensitive troponin assays are, indeed, more sensitive.  The sensitivity comes at a cost of specificity.  We are also encountering a sort of Zeno’s Paradox with regard to our evaluation of these strategies.  We employ more sensitive assays both to detect AMI up-front, but also as our gold-standard for adjudicated final outcome.  The increased sensitivity, then, cuts both ways – as we detect tinier and tinier nSTEMI as a sort of ceiling for our sensitivity of any rapid rule-out strategy.  Is it reasonable to suggest the occasional 1-in-200 tiny myocardial injury missed is unlikely to have serious clinical consequences?  If so, the best question to address is how to continue the care of these patients after they leave the Emergency Department such that those who would benefit from additional medical and endovascular intervention are not lost.

“The use of very low concentrations of high sensitivity troponin T to rule out acute myocardial infarction using a single blood test”
http://www.ncbi.nlm.nih.gov/pubmed/27178492

No Single Reliable Criterion for Diagnosing Acute Heart Failure

Some disease processes have been discretely boiled down to a single affirmative or exclusive test.  The positive or negative likelihood ratios associated with these tests can be applied to specific diagnoses almost absent any clinical context.

Acute heart failure is not one of these diagnoses.

This systematic review of clinical and laboratory features associated with a diagnosis of acute heart failure in the Emergency Department reveals the cupboard is mostly barren.  Most folks suggest a +LR >10 or a -LR <0.1 is a reasonable threshold for judging the informational value of a test.  These authors – in order to generate anything like a positive reflection of the evidence – were charitable enough to include anything with a +LR >4 or -LR of <0.25 as meaningful.

And, even that didn’t generate much:

  • An S3 on cardiac auscultation (+LR ~4.0).
  • Chest x-ray sigs of fluid overload (+LR ~4.8-6.5)
  • BNPs <100 or NT-proBNP <300 (-LR ~0.1)
  • BNPs >500 ranged in +LR from ~4-9, while NT-proBNP never exceeded a +LR of ~3.3.
  • Lung ultrasound probably had the best combination of LRs, with B-line scan having +LR ~7.4 and -LR ~0.16.
  • Specific findings on bedside echocardiography seem potentially valuable, but very few patients were included in these studies.

The short answer: acute heart failure remains a multifactorial evaluation, and it would be erroneous to routinely rely on any single test.

“Diagnosing Acute Heart Failure in the Emergency Department: A Systematic Review and Meta-analysis”
http://onlinelibrary.wiley.com/doi/10.1111/acem.12878/full

Let’s Get Inappropriate With AHA Guidelines

How do you hide bad science?  With meta-analyses, systematic reviews, and, the granddaddy of the them all, guidelines.  Guidelines have become so twisted over the recent history of medicine the Institute of Medicine had to release a statement on how to properly create them, and a handful of folks have even gone so far as to imply guidelines have become so untrustworthy a checklist is required for evaluation in order to protect patients.

Regardless, despite this new modern era, we have yet another guideline – this time from the American Heart Association – that deviates from our dignified ideals.  This guideline is meant to rate appropriate use of advanced imaging in all patients presenting to the Emergency Department with chest pain.  This includes, for their purposes, imaging to evaluate nSTEMI/ACS, suspected PE, suspected syndromes of the aorta, and “patients for whom a leading diagnosis is problematic or not possible”.

My irritation, as you might expect, comes at the expense of ACS and “leading diagnosis is problematic or not possible”.  The guidelines weighing the pros and cons of the various options for imaging PE and the aorta are inoffensive.  However, their evaluation of chest pain has one big winner: coronary CT angiograms.  The only time this test is not appropriate in a patient with potential ACS is when the patient has a STEMI.  They provide a wide range of broad clinical scenarios to assist the dutiful reader – all of which are CCTA territory – including as every low/intermediate risk nonischemic EKG and troponin-negative syndrome, explicitly even TIMI 0 patients.

Their justification of such includes citation of the big three – ACRIN-PA, ROMICAT II, and CT-STAT – showing the excellent negative predictive value of the test.  Indeed, the issues with the test – middling specificity inflicted upon low disease prevalence, increased downstream invasive angiography and revascularization of questionable value – are basically muttered under the breath of the authors.  Such dismissive treatment of the downsides of the test are of no surprise, considering Harold Litt, of ACRIN-PA and Siemens, is part of the writing panel for the guideline.  I will, again, point you to Rita Redberg’s excellent editorial in the New England Journal of Medicine, refuting the foundation of such wanton use of CCTA in the emergency evaluation of low-risk chest pain.

The “leading diagnosis is problematic or not possible” category is just baffling.  Are we really trying to enable clinicians to be so helpless as to say, “I don’t know!  Why think when I can scan?”  The so-called “triple rule-out” is endorsed in this document for this exact scenario – so you can use a test whose characteristics for detection of each entity under consideration are just as degraded as your clinical acumen.

Fantastically, both the Society of Academic Emergency Medicine and the American College of Emergency Physicians are somehow co-signatories to this document.  How can we possibly endorse such fragrant literature?

“2015 ACR/ACC/AHA/AATS/ACEP/ ASNC/NASCI/SAEM/SCCT/SCMR/ SCPC/SNMMI/STR/STS Appropriate Utilization of Cardiovascular Imaging in Emergency Department Patients With Chest Pain”
http://www.ncbi.nlm.nih.gov/pubmed/26809772

CTCA in the ED, Getting Less Sexy By The Day

As I’ve written before, the CT coronary angiograms is a funny test.  The idea of having a non-invasive method of detecting previously unknown coronary artery disease is compelling.  The practical application, however, has been limited by a low specificity – further exacerbated by those encouraging its use in a population with low pretest probability.

However, the few major studies regarding it tend to view CTCA in a favorable light – the result of comparing CTCA-based strategies to modern over-triage and over-testing of potential acute coronary syndrome.  These studies, ACRIN-PA and ROMICAT, showed significant improvements in direct discharge from the ED and in length-of-stay, not so much due to being a superior strategy of benefit to patients, but by obviating unnecessary care inflicted upon them.

The general gist of this trial is framed in the “era of high-sensitivity troponins” – referring to new developments in assays allowing a safer rapid rule-out in the Emergency Department.  This trial, as opposed to the others, also occurs in the Netherlands, a setting in which direct discharge from the ED is no anathema.  The “standard care” arms of ACRIN-PA and ROMICAT-2 had discharge rates from the ED of ~20% or less, while this trial discharges nearly 60%.  Yet, despite such recklessness displayed in this trial, these patients are ultimately just as safe.  And, when such an insanity-reduction initiative is undertaken, the advantages of CTCA diminish.

And, frankly, nearly all low-risk patients can be discharged safely from the Emergency Department.  The appropriate urgent follow-up test, if any, is a trickier proposition – and CCTA may yet be appropriate for some.  However, as a routine, ED-based strategy, it should probably be considered low-value care.

“Coronary CT Angiography for Suspected ACS in the Era of High-Sensitivity Troponins”
http://content.onlinejacc.org/article.aspx?articleID=2479881