What is the purpose of a pharmacist?

[kwiteaboy]

Originally Posted by bigasahouse

Computers are not better at reading EKG's than a cardiologist. They often read things that are not there. Every single EKG done in our hospital has to be read additionally by a cardiologist. You cannot rely on the computer's reading.

I understand that EKGs are always read by cardiologists. This article (published 19 (!) years ago) isn't the specific cite I was looking for, but it works: http://www.ncbi.nlm.nih.gov/pubmed/2233377

Computers in 1990 were "almost as accurate" at cardiologists in making seven major diagnoses. The cite I was thinking of was published within the last few years and found pretty strong evidence that modern computers are better than cardiologists at making diagnoses off of EKGs.

 

[bigasahouse]

Originally Posted by kwiteaboy

I understand that EKGs are always read by cardiologists. This article (published 19 (!) years ago) isn't the specific cite I was looking for, but it works: http://www.ncbi.nlm.nih.gov/pubmed/2233377

Computers in 1990 were "almost as accurate" at cardiologists in making seven major diagnoses. The cite I was thinking of was published within the last few years and found pretty strong evidence that modern computers are better than cardiologists at making diagnoses off of EKGs.

In straightforward EKG's, the computer does well. However, they have limitations when the EKG is more complex.

Originally Posted by Status of Computerized Electrocardiography
Cardiology Clinics - Volume 24, Issue 3 (August 2006)

The more straightforward diagnoses appear to be better detected by the computer, and most time is saved when the ECG is simple but tedious (multiple diagnoses) to read [35], [38]. The computerized reading also offers a real-time second opinion that may prevent "inadvertent oversight." When the diagnoses are ambiguous or controversial, computerized interpretations can make over reading more time consuming [35].

The anecdotal experience of many ECG readers is that a computer interpretation of "normal ECG" is usually, but not always, correct (Fig. 2). Poon and colleagues [39] found the computer to have high sensitivity of 98.7% (3531/3579), but somewhat lower specificity of 90.1% (338/375) in diagnosing sinus rhythm. The authors cautioned against routinely accepting a computer interpretation of "sinus rhythm" because close to 10% of nonsinus rhythms were erroneously interpreted as sinus rhythm. Because the prevalence of sinus rhythm was high in this study (3579/3954), however, the positive predictive value of a computer diagnosis of "sinus rhythm" was 99.0%, supporting the sentiment that a computer statement of "sinus rhythm" is generally reliable.

LIMITATIONS OF COMPUTERIZED ECG ANALYSIS
Limitations of electrocardiography as a diagnostic modality carry over to computerized electrocardiography. One of the most important limitations of electrocardiography is the inability to effectively detect certain anatomic and pathophysiologic conditions. Despite the development of multiple diagnostic ECG criteria, electrocardiography has only around a 50% sensitivity in detecting left ventricular hypertrophy that has been established by echocardiography (Fig. 3) [40], [41], [42]. It has been recognized for years that electrocardiography is better at diagnosing arrhythmias and conduction disturbances than structural cardiac or metabolic abnormalities. Ironically, rhythm interpretation is recognized to be one of the most difficult tasks for the computer.

Only a handful of studies have specifically evaluated the accuracy of computerized rhythm diagnosis. Shirataka and colleagues [24] found that although sinus rhythm with first-degree AV block had 100% agreement with the reference standard across five different computer programs, the accuracy of diagnosing second-degree AV block varied between 0% and 100%. In a study that assessed 11,610 consecutive computerized ECG interpretations, Varriale and colleagues [43] reported that multifocal atrial tachycardia was universally misdiagnosed as atrial fibrillation, accounting for 14% of ECGs interpreted as atrial fibrillation. Bogun and colleagues [27] found that of 2298 ECGs interpreted to be atrial fibrillation, 19% were misinterpreted by the computer. The difficulties in diagnosing atrial fibrillation is one of the most commonly encountered problems with computerized rhythm analysis (Fig. 4). More recently, Poon and colleagues [39] examined computerized rhythm interpretation by analyzing 4297 consecutive ECGs using one of the more advanced programs currently available (GE Marquette, version 19). The computerized interpretation was assessed against a consensus interpretation by reading cardiologists. Overall, 13.2% of the rhythm interpretations made by the computer were incorrect, and 7.8% if cardiac pacemaker activity misinterpretation was excluded.

The correct identification of pacemaker activity is still a major problem for today's computer software [44]. Poon and colleagues [39] reported that 75.2% of all pacemaker rhythms were misinterpreted. Because pacemaker stimulus outputs are very short in duration (generally 0.4-0.6 milliseconds), high sampling rates (>1000 Hz) are needed for adequate detection. In addition, current bipolar pacemaker systems are able to capture the myocardium with smaller amounts of energy, in turn resulting in progressively smaller pacemaker stimulus outputs that can be undetectable on the surface ECG. This reduction in stimulus output is facilitated by autocapture algorithms that allow the device to deliver its output just above the capture threshold, achieving pacemaker battery conservation while maintaining patient safety. Current algorithms "write in" the pacing output stimulus to make pacing activity apparent. High-frequency components of the ECG, such as pacemaker output stimuli, can sometimes disappear when the tracing is reprinted after data has been compressed for storage; the noncompressed ECG has the best signal fidelity and is the most appropriate for determining pacemaker rhythm. Considerable improvement has been made [45], however, especially in the detection of ventricular pacemaker activity, and newer iterations of the available algorithms are expected to have even further enhanced ability to define ventricular and atrial pacemaker stimuli.

There does appear to be progressive improvement in rhythm analysis software. The Marquette 12SL ECG Analysis software program (GE Health Care Technologies, Waukesha, Wisconsin) has been the single most studied commercial computer algorithm through its many versions [22], [23], [29], [31], [38], [39]. The most current version 20 has been found to need physician correction of a rhythm interpretation in only 4.1% of cases [44] compared with 7.8% reported with version 19 [39]. Although interpretive software from major manufacturers, such as GE Marquette and HP Phillips, include more sophisticated programming that specifically address difficult aspects of rhythm analysis, such as pacemaker activity recognition and analysis of pediatric ECGs, continued development is still needed [38].

 

[kwiteaboy]

Interesting - thanks for sharing that.

 

[Milhouse]

Just out of curiosity, how accurate are physicians at reading ECGs?

 

[apropos]

Originally Posted by Milhouse

Just out of curiosity, how accurate are physicians at reading ECGs?
Like most medicine-related things, variable, and it depends on their exposure/experience.

 

[thebeatblitz]

Originally Posted by kwiteaboy

Interesting - thanks for sharing that.

Anyone constantly reading them (EM, IM, Cards, CT, Gas) is probably pretty good. It takes lots of practice to get good at it, and most other specialties are only peripherally exposed to ECGs aside from PGY1.

 

[Milhouse]

Originally Posted by apropos

Like most medicine-related things, variable, and it depends on their exposure/experience.

So, when discussing the value of computer analysis of ECG. . . what isn't important is how precise and accurate the machine is on its own. . .but rather how much better or worse it is than the average physician reading the ECG. A comparison.

Is there a better outcome by using a computer instead of a physician? By using a physician instead of a computer?

Or, most likely, using a computer AND a physician with a set of SOPs.

 

[thebeatblitz]

The "average" physician isn't going to be reading the ECG of a complicated CV patient. If someone comes into the ED with chest pain and has an abnormal ECG, Cards will be consulted. Machine reads are already used in emergent situations; most new AEDs will only shock if a treatable rhythm is detected. There is a ton of money being poured into developing better automated this-and-that, but it's all relatively meaningless unless you have a proper history and exam to go along with it. I'm matching into a field dominated by technology (Rads), but without a clinical picture there really isn't much we can differentiate. In medicine, there are only a handful of pathognomonic conditions. Until we have a robot that can do the history and exam and be able to tell which patients aren't being truthful, automation only goes so far. And the lawyers need someone to sue; you can't sue a machine for a bad read.

 

[Milhouse]

Originally Posted by thebeatblitz

The "average" physician isn't going to be reading the ECG of a complicated CV patient. If someone comes into the ED with chest pain and has an abnormal ECG, Cards will be consulted.

Machine reads are already used in emergent situations; most new AEDs will only shock if a treatable rhythm is detected.

There is a ton of money being poured into developing better automated this-and-that, but it's all relatively meaningless unless you have a proper history and exam to go along with it. I'm matching into a field dominated by technology (Rads), but without a clinical picture there really isn't much we can differentiate. In medicine, there are only a handful of pathognomonic conditions.

Until we have a robot that can do the history and exam and be able to tell which patients aren't being truthful, automation only goes so far.

And the lawyers need someone to sue; you can't sue a machine for a bad read.

Cute, but you skipped a part of my quote "the average physician READING THE ECG".

 

[thebeatblitz]

What do you mean? There is no such thing as an "average physician" reading an ECG. There are tons of physicians in all sorts of specialties. People that are trained to read ECGs read ECGs. The pulmonologist is not going to come in an read an ECG when someone suspects PE because he feels like it. To be specific: The average cardiologist will probably read just as well as a machine in an uncomplicated situation. In complicated situations, I would trust the average cardiologist more than a machine. Think about it: How do machines "know" what to detect? They have pre-programmed algorithms to detect abnormalities against a normal QRS. Unfortunately, the medical literature changes monthly, which means a machine's software would have to be updated monthly. So, only the most uncomplicated rhythms can be programmed into a machine to be completely reliable. This is already done in AEDs; they won't shock asystole or PEA, which almost anyone can see (flatline). Most monitors start sirening when they detect abnormalities. Is this what you mean by marrying the two? So, yes. If you take a podiatrist and ask him to read an ECG, I would trust the machine more. Of course, this never happens so it's comparing apples to oranges. If you take a cardiologist, specifically an electrophysiologist, then I would take him over a machine. I'm sure there have been studies on this because there always are in technology-driven fields.

 

[Milhouse]

Average refers to the mean. It is a statistics word. Physician refers to a person that holds an MD. Reading an ECG means interpreting the electrocardiogram.

You see, we can design something called an experiment. We know we have a population of physicians that read ECGs as a major responsibility. We can sample those physicians' interpretations of ECGs. We can compare those interpretations to samples of how a machine interpreted the ECG. We can then determine which results in better outcomes for the patient.

 

[Milhouse]

BTW, this is why MDs need to stay out of science. I can't even begin to count the number of times I have had to tell an MD "no, you didn't prove anything" or "you never used proper controls, your results are meaningless".

 

[kalice]

Originally Posted by Milhouse

BTW, this is why MDs need to stay out of science. I can't even begin to count the number of times I have had to tell an MD "no, you didn't prove anything" or "you never used proper controls, your results are meaningless".

LOL. I'm sorry, but this is far from the truth. MD's actually contribute a huge amount to basic and clinical science, and MD PhD's even more so. It's unlikely the run of the mill GP or family physican without the NIH grants or access to resources will be studying the role of adherens junction tension in chondrocyte migration across different substrates to provide a framework for cartilage regrowth in osteoarthritis. However, all physicians are trained, at least in the first two years of school, in basic medical science (in addition to pre-req's in undergrad), and in additional, many residencies and fellowships have build in either clinical or basic research time as part of the training.

However, in defense of career tracks, research or instructor track physicans who are intimately involved in medical research or teaching are usually concentrated at large teaching institution. And most physicans who are either private, non-teaching, or just simply do not keep up with the most up to date research will have a less firm grasp on the cutting edge stuff.

Medicine is a constantly evolving field, and the standards of care, as well as which medication to choose under what parameters is constantly changing, therefore a physican has to be adept at understanding the clinical trials and experiments and make decision based on the results. They need to be trained (and are) in understanding experimental setup, controls, and decide whether the outcomes are valid or not, in order to make decisions that is best for the patient. Not all the decisions are cookie cutter alogrithims that was taught in school or empirically based on past experience.

 

[Milhouse]

I was probably a bit harsh with the last comment, but there are just so many asshole MDs out there that think they know everything about science that it gets really frustrating to work with them, because they somehow can't or won't see that they are wrong.

Science and statistics are both skills that if not used constantly, end up forgotten. Even scientists regularly consult statisticians on projects. But many of these MDs I've worked with seem to think that they don't need anyone else to help them.

A lot of this is why I hate what I do for a living.

 

[kwiteaboy]

Originally Posted by thebeatblitz

And the lawyers need someone to sue; you can't sue a machine for a bad read.

This is the major reason computers probably won't ever replace physicians in this respect.