Vital Statistics Blog

Publication in the Age of AI

2026-07-02T00:00:00.000Z

Scientific publication is being overwhelmed by artificial intelligence. We have entered an age in which it is faster to write a paper than it is to properly review it. An artificial intelligence (AI) agent can search a public dataset for every pairwise relationship, keep the few that clear significance, and draft a manuscript almost before a referee has time to read the abstract. The cost of producing a plausible finding – backed by real data – is rapidly approaching zero while the cost of evaluation remains unchanged. This asymmetry threatens to flatten the signal-to-noise ratio in scientific literature.

The challenge here is not fraud; these (AI) authors are genuinely running reasonable analyses on real data. The issue is that these analyses can be run on a massive scale. Exploratory work like this can be meaningful, but only when replicated in new data, or when integrated with related findings.

More than the narrative

We believe that AI can be used as a mechanism to alleviate this AI-generated problem. However, it will require restructuring what it means to publish. We need to exploit AI for its strengths while ensuring it isn't used where it isn't trustworthy.

We propose placing data and its metadata at the center of publication. A complete research product should include the dataset (or the means to compute against it), the metadata needed to understand and reuse it, the analysis code with the details of the computing environment, and an interpretation or narrative that is understandable by people and ingestible by machines. We recognize that this is not a new idea – consider the research compendium and the FAIR (Findability, Accessibility, Interoperability, and Reusability) standard. However, AI makes this achievable in a way that is far less prescriptive (and restrictive) than ever before.

By taking advantage of the ability of AI to translate natural language plus metadata into analysis, we can construct a computing environment that facilitates the publication of robust research products while allowing data of essentially arbitrary structure to be both protected and broadly usable (Figure 1). This environment is:

Secure – isolates the data in a secure environment (if so desired)
Auditable – enables vetting by experts to guard against hallucination
Flexible – can be used with essentially any data layout; allows arbitrary analyses
Version controlled – technical bookkeeping such as versioning of data, computing environment, scripts, and chat logs is automatic
Reviewer friendly – allows reviewers to automatically verify analyses through the direct submission of analysis scripts
Transparent – the entire arc of an analysis can be tracked through examination of the agent chat logs

Figure 1 – Protecting data while allowing computation. AI agents write code against metadata, but never have direct access. Results are vetted to ensure no PHI is released. We recognize that no system is perfect, but isolating the data from the requester, together with standard data use agreements and legal protections, can support a robust, generally accessible data environment.

As these environments spread, it becomes possible to distribute analyses across many datasets. This allows early vetting of biological hypotheses based on what is already available, and focused experiments to fill in missing pieces. The idea of distributing analyses across many datasets is also not new. Consider the OMOP (Observational Medical Outcomes Partnership) data standard used by OHDSI (Observational Health Data Sciences and Informatics) to run the same analyses in a distributed way across many databases. The flexibility of AI allows federated management of data and compute environments as well as distributed analysis across much more disparate datasets.

Data marketplaces

Incentives, not technology, will decide whether data gets shared and how research publication changes. The institutions that hold the most valuable data – health systems, pharmaceutical companies, biobanks – have no reason to give it away. Asking them to do so is futile and devalues their contributions.

The publication model we propose lets the data stay where it is. The holder (i) controls access, (ii) publishes metadata rich enough for an outside analyst, or an AI, to write an analysis, and (iii) runs that analysis inside its own enclave and returns the result. Compute travels to the data so that the data is never out in the open. When governed appropriately to avoid re-identification, this converts data you cannot share into both a financial asset for the owner and a research asset for the community.

This approach ensures that data sharing does not become a tax on the large players; it is instead an instrument that incentivizes broader collaboration and data access. Building a clean, well-documented dataset becomes financially viable. Analyses spanning many modalities – claims, genomics, imaging, clinical notes – become possible, and each holder contributes without losing control.

Peer review

Vetting research products looks different from reviewing articles. Reproducing figures and tables is mechanical and automatic; this allows reviewers to spend their effort where judgment is required.

An AI "reviewer" can:

Search for relevant public datasets the authors missed and run confirmatory analyses
Directly test whether results are robust to reasonable analytic variation (see multiverse or specification curve analysis)
Determine whether the metadata suffices to link the data to other published datasets

Even though much of the validation can be done by the AI, there are questions that aren't answered by reproduction. These are the more complex, subjective questions for which human reviewers are better suited.

Do the statistics support the written claims?
Does the arc of exploration (available in the chat logs with the AI) warrant a strong or a weak conclusion?
Is the narrative accurate, and does it matter to patients, medicine, and/or society?

Conclusion

When a complete research product is published, our entire scientific literature changes shape. Null results do not require a special venue; they are simply available. Findings are no longer single events; they instead transform into a living record represented by evidence that updates as data arrives. Meta-analyses are available on demand and with statistical backing, facilitating rapid understanding of the cutting edge of research in any field.

Most importantly, much of the rote part of the work associated with peer review can be automated. Widely shared data, carrying the metadata needed to use it, opens approaches to discovery that we have not yet imagined.

Disruption Through Specialization

2015-04-22T00:00:00.000Z

It has been hypothesized that disruptions in medical care will be partly driven by the commoditization of certain procedures. While hospitals can address almost any ailment, their level of expertise and flexibility comes with a significant overhead cost. There are many diseases that can be diagnosed and treated without need for the wide range of expertise available in hospitals - consider for example the array of services offered by nurses in retail clinics like [Target](http://www.target.com/c/clinic-services-health/-/N-4yia4#?lnk=lnav_ clinic_1&intc=2413603|null)and Walmart. For health systems that share some of the cost when patients get sick, efficient care delivery is quickly becoming critically important.

One of the most effective methods to lower health system costs is the creation of specialty clinics. These clinics save money by streamlining patient flow for specific treatments and moving part of the burden of administering treatment to skilled but lower cost staff. Additional benefits of this approach to care delivery include a greater level of consistency and clear performance metrics that are relevant to the specific therapy being delivered.

While the benefits of moving patients out of high-cost hospital settings and into lower cost specialty clinics are significant, it isn't obvious which patients and procedures are best suited to this approach. In this article we examine whether the 2012 Medicare Part B claims data can help us find providers that have focused their practice for the delivery of specific therapies, whatever the reason for that focus.

Proceduralists. There are a little over 46,000 proceduralistsin the Medicare data set; these are healthcare providers earning greater than 90% of their Medicare income from repeated applications of a single procedure. Figure 1 compares the number of procedures performed to total dollars earned from that procedure for all physicians in the data set and highlights similar results for proceduralists.

Figure 1 - Favorite Procedures. Each spot corresponds to a single care giver in the data set and shows the count and income from the single most performed procedure for that provider. Providers that earn greater than 90% of their Medicare income from a single procedure are colored in red. There is an obvious predilection for proceduralists to focus on more expensive therapies as seen by the upward shift in the red points compared to the gray.

Approximately half of the proceduralists are chiropractors with an additional 12,000 social workers, clinical psychologists and psychiatrists making up the next most common types of healthcare provider to specialize in just one procedure. Providers in these categories appear to be fundamentally different from other providers in multiple ways; for example, they are also among the least likely to see Medicare patients or to "accept assignment" when they do take Medicare patients. Because there are so many of these practitioners, the procedures they focus on tend to receive the largest total dollars from Medicare (Table 1) even though the amount earned by each practitioner is not particularly high.

Hemophelia. One of the procedures with high total expenditure that is not explained by the large number of practitioners is "Factor VIII recombinant NOS". This has been the focus of a previous article on opportunities to lower Medicare costs by changing incentive patterns. That article pointed out the opportunity to save an average of $10,000/month per patient by moving those patients to the non-recombinant version of Factor VIII. There are only 8 Factor VIII proceduralists; nonetheless there is the opportunity to save Medicare multiple millions of dollars by modifying the practice behavior of just these 8. (In fact, there are only 5 since Accredo Health Group Inc. accounts for four locations in California, North Carolina, Florida and New Jersey). We note that there are no proceduralists focused on the treatment of factor VIII deficiency who are using the lower cost version of this medication - presumably because of the backwards financial incentive structure built into the Medicare fee schedule.

Home INR monitoring. Among proceduralists, code G0249 is the single procedure that accounts for the highest income per provider (first line of Table 2). This is intended for providers of kits for in-home monitoring of blood clotting; monitoring of this type is typically intended for patients who are on blood thinners and need to ensure that they are not over or under medicating. While the average income of $10 million per lab that specializes in this procedure sounds exorbitant, these 8 labs manage home monitoring for over 120,000 patients. Because of the high levels of morbidity and mortality associated with poorly managed blood thinners, $650/year per patient spent on home monitoring is probably an excellent investment.

Drugs. There are a number of proceduralists making over $100,000 (gross) from the administration of particular drugs. These include the cancer drugs Ipilimumab and Bevacizumab, drugs for immune deficiency (Octagam, Gamunex, Privigen, Immune Globulin), Natalizumab for Crohn's disease and Multiple Sclerosis, Imuglucerase for Gaucher's Disease, Infliximab for Rheumatoid Arthritis and Crohn's disease and Omalizumab for asthma. Each of these drugs has from 1 to 3 proceduralists focusing on their therapeutic use with the exception of Infliximab and Natalizumab (8 and 11 proceduralists respectively).

It is difficult to believe that differences in patient population can explain the presence/absence of proceduralists in different markets. For example, 4 of the 11 Natalizumab proceduralists operate in the same practice in Raleigh, NC, and the only Octagam proceduralist practices in Irving, Texas. If we do not believe that there is a disproportionate percent of the population in Irving with immune deficiency, we must conclude that either (1) building a practice that focuses on this procedure has increased the efficiency of treating this population or that (2) some patients in Irving are being overtreated with Octagam.

Conclusion. There are a total of 332 procedures for which there exists at least one proceduralist provider. Examples of billable procedures, other that the few we have already mentioned, include miles traveled for the collection of blood to be used to in laboratory tests, movement of portable x-ray machines, management of end stage renal disease, various examinations of tissue samples by pathologists and others.

Without more extensive research, it is difficult to determine whether any particular proceduralist is boosting the efficiency of the healthcare delivery in the systems in which they work or whether they are overusing their procedure of choice. In either case, there is tremendous opportunity to boost the efficiency of healthcare delivery by identifying these practitioners and either emulating them more broadly or encouraging them to modify their practices.

Table 1 - Specialty procedures sorted by the total dollars earned by practitioners focused on those procedures.

Code	Description	Cost per procedure	# Specialists	Total $	Average Income
98941	Chiropractic manipulation	$34.30	14433	$286,515,905	$19,851.44
90806	Psytx off 45-50 min	$72.00	8031	$167,043,144	$20,799.79
88305	Tissue exam by pathologist	$70.20	544	$165,598,150	$304,408.36
99214	Office/outpatient visit est	$83.00	2436	$84,513,755	$34,693.66
98940	Chiropractic manipulation	$25.20	6341	$84,096,788	$13,262.39
G0249	Provide INR test mater/equip	$119.00	7	$73,450,450	$10,492,921.43
99213	Office/outpatient visit est	$56.80	2875	$73,368,938	$25,519.63
84999	Clinical chemistry test	$1,160.00	6	$58,602,711	$9,767,118.50
90862	Medication management	$55.40	1946	$49,367,138	$25,368.52
J7192	Factor viii recombinant NOS	$12,500.00	8	$36,599,248	$4,574,906.00
98942	Chiropractic manipulation	$44.20	1410	$33,290,713	$23,610.43
93229	Remote 30 day ecg tech supp	$650.00	6	$30,048,559	$5,008,093.17
97110	Therapeutic exercises	$28.10	1073	$28,860,561	$26,897.07
90807	Psytx off 45-50 min w/e&m	$100.00	894	$28,161,756	$31,500.85
90960	Esrd srv 4 visits p mo 20+	$259.00	175	$22,660,952	$129,491.15
J3490	Drugs unclassified injection	$678.00	10	$15,690,592	$1,569,059.20
90805	Psytx off 20-30 min w/e&m	$69.10	640	$14,703,880	$22,974.81
90818	Psytx hosp 45-50 min	$67.10	239	$13,985,234	$58,515.62
99215	Office/outpatient visit est	$126.00	308	$12,285,748	$39,888.79
64450	N block other peripheral	$53.30	25	$10,980,400	$439,216.00
86849	Immunology procedure	$2,820.00	1	$10,145,033	$10,145,033.00
90808	Psytx office 75-80 min	$103.00	248	$10,012,582	$40,373.31
90804	Psytx office 20-30 min	$53.70	219	$7,072,871	$32,296.21

Table 2 - Specialty procedures sorted by the average income of providers focused on those procedures. Note that this representation may be misleading as there are instances in which all providers within a practice are aggregated into a single provider in the Medicare data.

Code	Description	Cost per procedure	# Specialists	Total $	Average Income
G0249	Provide INR test mater/equip	$119.00	7	$73,450,450	$10,492,921.43
86849	Immunology procedure	$2,820.00	1	$10,145,033	$10,145,033.00
84999	Clinical chemistry test	$1,160.00	6	$58,602,711	$9,767,118.50
93229	Remote 30 day ecg tech supp	$650.00	6	$30,048,559	$5,008,093.17
J7192	Factor viii recombinant NOS	$12,500.00	8	$36,599,248	$4,574,906.00
J9228	Ipilimumab injection	$103.00	2	$5,524,475	$2,762,237.50
L9900	O&P supply/accessory/service	$1,720.00	2	$4,273,886	$2,136,943.00
0297T	Ext ecg scan w/report	$263.00	1	$1,635,151	$1,635,151.00
J3490	Drugs unclassified injection	$678.00	10	$15,690,592	$1,569,059.20
J1786	Imuglucerase injection	$41.30	3	$4,682,398	$1,560,799.33
G0166	Extrnl counterpulse, per tx	$141.00	3	$3,231,791	$1,077,263.67
J1561	Gamunex/gamunex c	$37.60	1	$1,003,937	$1,003,937.00
J9035	Bevacizumab injection	$55.80	3	$2,727,958	$909,319.33
83901	Molecule nucleic ampli addon	$19.70	1	$757,658	$757,658.00
96102	Psycho testing by technician	$73.00	1	$650,552	$650,552.00
G9152	Mapcp demo community	$4.80	8	$4,958,309	$619,788.63
J1568	Octagam injection	$37.90	1	$543,460	$543,460.00
J2323	Natalizumab injection	$11.20	11	$5,761,235	$523,748.64
64450	N block other peripheral	$53.30	25	$10,980,400	$439,216.00
J1459	Inj IVIG privigen 500 mg	$32.00	2	$859,256	$429,628.00
46500	Injection into hemorrhoid(s)	$210.00	1	$415,388	$415,388.00
36516	Apheresis selective	$571.00	1	$406,277	$406,277.00
93293	Pm phone r-strip device eval	$48.00	9	$3,106,620	$345,180.00
J1745	Infliximab injection	$59.40	8	$2,747,242	$343,405.25
17004	Destroy premal lesions 15/>	$167.00	2	$678,976	$339,488.00
88305	Tissue exam by pathologist	$70.20	544	$165,598,150	$304,408.36
93271	Ecg/monitoring and analysis	$135.00	1	$301,946	$301,946.00
99183	Hyperbaric oxygen therapy	$202.00	9	$2,437,519	$270,835.44
G0399	Home sleep test/type 3 Porta	$162.00	3	$776,218	$258,739.33
90802	Intac psy dx interview	$139.00	1	$236,990	$236,990.00
36478	Endovenous laser 1st vein	$1,200.00	4	$935,864	$233,966.00
37227	Fem/popl revasc stnt & ather	$2,280.00	1	$230,344	$230,344.00
J1566	Immune globulin, powder	$28.70	1	$224,439	$224,439.00
P9603	One-way allow prorated miles	$0.98	7	$1,521,184	$217,312.00
36475	Endovenous rf 1st vein	$760.00	8	$1,656,120	$207,015.00
88348	Electron microscopy	$569.00	1	$190,451	$190,451.00
77522	Proton trmt simple w/comp	$784.00	1	$181,031	$181,031.00
95951	EEG monitoring/videorecord	$996.00	4	$710,637	$177,659.25
90814	Intac psytx off 75-80 min	$100.00	2	$338,108	$169,054.00
86481	Tb ag response t-cell susp	$97.00	2	$334,498	$167,249.00
G0431	Drug screen multiple class	$78.00	15	$2,465,313	$164,354.20
J2357	Omalizumab injection	$22.20	3	$443,261	$147,753.67
88304	Tissue exam by pathologist	$62.70	2	$291,353	$145,676.50
L8680	Implt neurostim elctr each	$300.00	3	$411,375	$137,125.00
88346	Immunofluorescent study	$94.00	1	$131,942	$131,942.00
90960	Esrd srv 4 visits p mo 20+	$259.00	175	$22,660,952	$129,491.15
86353	Lymphocyte transformation	$47.60	1	$119,647	$119,647.00
17108	Destruction of skin lesions	$618.00	1	$109,377	$109,377.00
93306	Tte w/doppler complete	$151.00	42	$4,576,030	$108,953.10
95811	Polysomnography w/cpap	$483.00	1	$104,329	$104,329.00
11042	Deb subq tissue 20 sq cm/<	$79.70	4	$403,377	$100,844.25

Approach. We identify proceduralists based on the percent of gross allowed payments from Medicare assigned to a certain procedure compared to total gross allowed Medicare payments. Providers who derive greater than 90% of their gross income from a single procedure are designated proceduralists. We look only for providers who have a total of at least 100 procedures recorded in the data. Because of the way the data are filtered, physicians who are eliminated by this criterion are unlikely to be specialists unless they work almost exclusively with patients under the age of 65 (ie. not covered by Medicare).

Originally published on EHR Analytics, 2015.

Availability of specialists for patients on Medicare

2015-03-04T00:00:00.000Z

For patients who are new to Medicare it can be difficult to find a physician. This barrier to healthcare is almost certainly stressful and may even be a source of additional morbidity and mortality.

Many physicians have entered into an agreement with Medicare whereby they agree to accept in payment no more than the allowed amount for their services; this is called accepting assignment. Based on this explanation, physicians who do not accept assignment are paid only 76 percent of the allowed amount by Medicare and are allowed to charge 115 percent of the allowed amount. Thus, patients are responsible for around 38 percent of the (somewhat larger) bill for services when they see physicians who take Medicare but have not accepted assignment (see Table 1).

Table 1. Comparison of costs between physicians who accept assignment and those who do not.

We are interested in identifying those physicians who are generally unwilling to accept the Medicare allowed amount - or unwilling to accept Medicare at all - for their services. Obtaining treatment from physicians in those specialties will generally be more difficult and more expensive for seniors.

Relationship between specialties and rates of acceptance

Digging into the 2012 CMS data we find that all except 1574 of the 880,644 providers in the database accept assignment. This corresponds to an overall rate of acceptance of 99.8 percent. However, this rate does vary for some specialties.

Included in the data for every provider are both therapeutic specialty and an indicator of whether the provider accepts assignment. We can check whether the fraction of providers who accept assignment varies by therapeutic area. If there are particular types of providers who are significantly less (or more) likely to accept assignment then we can assume that the allowed payment amounts for procedures performed by those providers are lower (or higher) than the healthcare market would support.

Figure 1. Probability of accepting assignment by specialty. The graph shows a subset of specialties that are statistically different from the baseline acceptance rate.

Based on Figure 1, chiropractors in the data set will accept assignment at a rate that is close to 16 times less than the general population. There are some clinical trials demonstrating that chiropractic therapies compare favorably to muscle relaxers for the treatment of lower back pain and neck pain. However, the evidence supporting chiropractic manipulation is controversial at best. In addition, there is the potential for serious adverse events such as stroke caused by tearing of the vertebral artery.

Other specialists that are particularly unlikely (relative to baseline) to accept assignment are psychiatrists, oral and maxillofacial surgeons, optometrists and dermatologists. However, while the relative rates of accepting assignment vary, the absolute probability of accepting assignment conditional on accepting Medicare at all is still quite high – around 97.5 percent for Chiropractors. Thus it is generally true that physicians who accept any payments from Medicare are very likely to accept assignment.

However, there is no legal compulsion for physicians to work with Medicare. How many providers are not listed in the Medicare data release because they are never paid by Medicare?

Specialists who are less likely to see any Medicare patients

The physician’s specialty book published by the Association of American Medical Colleges lists physicians by specialty. Included in that census are 656,000 physicians broken into 36 different specialties. For comparison, there are 560,000 unique physicians (healthcare workers with either an MD or DO) in the CMS data set. If there are particular specialties in which physicians are systematically avoiding (or seeking out) Medicare patients then the proportion of physicians in those specialties relative to the group will be different between Medicare and the overall physician census. For example, the census data list 7,706 physicians who specialize in Child Psychiatry. Considering the age restriction associated with enrolling for Medicare, it shouldn’t be a surprise to learn that there are no Child Psychiatrists listed in the CMS data.

Figure 2. Number of physicians who take payments from Medicare compared to a general census of physicians conducted by the Association of American Medical Colleges. Physicians specializing in pediatrics or preventive medicine are under-represented among those taking payments from Medicare. Specialists in ophthalmology and emergency medicine are over-represented. Chiropractors are not included in this graph because they are not included in the census data.

Figure 2 shows how estimates of the number of physicians in each specialty differs when conducted with these two disparate data sets. Notice that there are some elements of this figure that suggest that the analysis makes some sense. For example, because Medicare patients are older one would expect there to be far more pediatricians in the census population – an assumption that is clearly born out in the data. In the opposite direction, emergency physicians are not allowed to turn away any patient with a medical emergency. As a result one would expect almost every emergency physician to take Medicare – the alternative would be to bill elderly patients for the full cost of emergency care (this result is also validated in our original analysis on rates of acceptance, see Figure 1).

Based on this analysis we see that ophthalmologists are more heavily represented in the Medicare physician population than they are in the general census. This may indicate that diseases of the eye are more prevalent in the elderly. On the other hand, a report in the New York Times in April of 2014 on a physician in south Florida who was paid $21 million by Medicare in 2012 would suggest that there may be a strong financial incentive for ophthalmologists to accept Medicare patients.

Based on both this analysis and on the analysis of acceptance rates (Figure 1), the Medicare fee schedule does not appear to be attractive to psychiatrists. It is clear that psychiatric illness places a financial burden on our healthcare system. What is somewhat more controversial is deciding what can be done about it. Despite tremendous ongoing effort to study the efficacy of psychiatry and psychiatric medications, there is a baseline mistrust of the field and a social stigma associated with receiving psychiatric care. Perhaps this also influences payment decisions made by CMS.

Results

The decisions that Medicare makes have a tremendous effect on the types of therapies that are made available to seniors. By choosing to put financial pressures on procedures that are most typically performed by certain physician specialists Medicare is decreasing the availability of those procedures. For the specialties in question, it may be that further proof that they benefit patient health is required for increased acceptance.

Originally published on EHR Analytics, 2015.

Procedure Costs and Treatment Decisions

2014-12-29T00:00:00.000Z

There have been a number of articles (1,2,3) discussing the financial incentives built into the healthcare system in the United States. Built into these discussions are two key assumptions: (1) that financial incentives are influencing treatment decisions and (2) that changing those financial incentives will lead to substantial cost savings for the same quality of care. The recent release of Medicare Part B payment data from 2012 has made available some anecdotal evidence that these two key assumptions are in fact correct. With this data it is possible to identify both the level of use and cost of individual therapies. Here are some examples of inefficiencies that are built into the way that healthcare decisions are incentivized in the USA.

Example: Hemophilia. The treatment with the highest minimum annual out-of-pocket expenses from the CMS data is “Factor VIII recombinant NOS”. This is a replacement for a missing protein in the blood of patients with Hemophilia A (also used to treat Von Willebrand disease). The disease causes spontaneous bleeding and according to the World Hemophilia Foundation children with severe, untreated Hemophilia A typically do not survive to adulthood. However, with treatment life expectancy is close to normal. Based on the CMS data table, the average cost to treat a senior with this condition is nearly $250,000/year with Medicare covering almost $200,000. In addition, the treatment is not a cure; it must be continued for the life of the patient.

In the portion of Medicare Part B that we can see, over $70 million – about $0.62 for every working American – was spent treating 357 seniors with Factor VIII recombinant NOS; those 357 seniors themselves were responsible for around $18 million in total costs.

There are other first line therapies for the treatment of Hemophilia A. In particular there is a non-recombinant version of factor VIII that is derived from blood plasma (also listed in the table). The major difference in these two products is the chance of transmission of communicable disease through the use of plasma products. However, due to modern techniques for detecting HIV and Hepatitis as well as pasteurization techniques, “no blood-borne transmission of hepatitis viruses or HIV has occurred in the last 20 years." Nonetheless, there are only 55 patients (13 percent of those treated with factor VIII or factor VIII recombinant) who are receiving plasma based factor VIII. If CMS had a mechanism for moving the patients on recombinant factor VIII to the non-recombinant version, it would lower its own costs by at least $34 million per year - $68 million if we extrapolate to the full size of Part B.

Example: Rheumatoid Arthritis. There is a similar comparison that can be made for two of the rheumatoid arthritis (RA) drugs listed in the "expensive treatments" table, and while the price differences are not nearly so dramatic, the number of patients affected is much larger. Infliximab ($13,430.70/year) and Tocilizumab ($10,997.55/year) are two competing therapeutics for RA. The two have been compared head to head in a clinical trial with Tocilizumab either equivalent or outperforming Infliximab on all measures of disease progression. Nonetheless, Infliximab was given to 42,645 patients in 2012 while Tocilizumab was given to only 2216. This comparison is not perfect because Infliximab is also an approved treatment for inflammatory bowel disease (IBD). However, the prevalence of IBD among seniors is much lower than that of RA. Even if we suppose that only a quarter of the Infliximab patients could be moved to Tocilizumab, Medicare would have saved $26 million on just this change in 2012.

Example: Prostate Cancer. Many of the most expensive treatments on the list are cancer therapies. Sipuleucel-T auto CD54+ is a treatment for end-stage, hormone refractory prostate cancer. Based on the three clinical trials that have been run using the treatment, it adds an average of about 4 months to the lifespans of treated patients. With the relatively high cost, it is likely that Sipuleucel-T auto CD54+ is inaccessible to most of the population of Medicare patients. If we estimate the prevalence of late stage prostate cancer based on statistics from cancer.org concerning death rates, then the problem afflicts around 30,000 men per year.

How do we, as a society, value those four months? Is it obvious that all 30,000 men with late stage prostate cancer be treated with Sipuleucel? This would require that every working American contribute approximately $2/year. Is it obvious that they shouldn’t? Similarly, it is difficult to deny factor VIII replacement to the 412 seniors in our data set who needed treatment in 2012, but do we as a society need to pay double for the recombinant version?

Perhaps even more important than these questions is this one: if patients are responsible for 20% of the cost of their therapies, why aren’t they demanding less expensive therapies when they are equally effective? The answer to this is probably very simple; Americans – and this sometimes includes physicians – simply don’t understand or think about the financial consequences of medical decisions. Very few people realize that if you order “J7190” instead of “J7192” from the menu of therapeutics (see the table), you get the same result, but your medical bill will drop by almost $10,000 per month.

Of the 4,309 different procedures that are commonly practiced and for which Medicare made at least a partial payment under Part B in 2012, 37 of them cost more than $10,000 per year. Of these procedures, 31 of them are medications. Currently, CMS is legally banned from bargaining with pharmaceutical companies for the price of medications. Instead it is bound on one side by the prices those companies set and on the other by the working definition of medically necessary. Physicians and pharma companies are generally making the purchasing and pricing decisions and both of them benefit from higher prices!

One solution to this problem is to demand that prices for medications be set in some way other than asking the companies who produce them how much they would like us to pay. On the other hand, letting the manufacturer set the price is the way that a free market society works. Another option is to change the incentive structure for the buyers. Under Part B, physicians generally get a percent of the cost back in profit. What would happen if this profit were unrelated to the cost of therapy or even inverted? Alternatively, perhaps the best solution is to ensure that the patients understand how much they are spending and what they are getting for their – and taxpayer – money.

Originally published on EHR Analytics, 2014.

Billing and financial incentives in healthcare

2014-12-06T00:00:00.000Z

In order to bill the insurance company, your physician must boil down each interaction and procedure performed into one of around 7,400 medical codes. In essence, when you seek medical care you are actually ordering from a precisely defined menu of around 7,400 dishes – even more than Cheesecake Factory. Important points to keep in mind: (1) no restaurant (clinic) serves every dish, (2) the prices are hidden, (3) the waiter (physician) has a very strong influence – almost complete in many cases – on what you order and (4) the tip is automatically included on the tab. If the analogy were complete, then you might expect to order a very nice bottle of wine every time you go out to eat.

To prevent people getting sick from all that wine, physicians have developed a society built around a very particular set of social norms, dating back at least to Hippocrates, that encourage priorities that are more in line with the patients’ needs. As in any society, the strength of influence of social norms varies from individual to individual, but the combination of norms, legal penalties and rewards tends to work for most.

If we generally trust physicians to keep the interest of their patients paramount, the next question becomes, how are those hidden prices set?

Medicare payments for Part B

Medicare uses a criteria for setting prices for each of the 7,400 therapies that can be described as “reimbursement for costs” (see this fact sheet for a description). The overall strategy is to compute three types of Relative Value Units (RVUs) – modified for cost of living – for each procedure that is reimbursed. These are associated with (1) the cost of performing the procedure, (2) the cost of maintaining a medical practice and (3) the cost of malpractice insurance. Any provider who has “accepted assignment” is required by this agreement to accept the agreed upon dollar amount for their services. There is an important caveat to this; physicians may accept assignment from Medicare but then refuse to see any patients who have Medicare as their primary insurance.

Medicare is supposed to pay for 80 percent of the allowed amount, but as can be seen from Figure 1 showing payments for Pemetrexed (a cancer medication), this percentage can vary. The usual reasons for this variation are supplemental insurance and deductibles.

There are two obvious difficulties with this approach to defining payments for medical services. First, there is the challenge of deciding who will estimate RVUs for each procedure. This is the task of the Relative Value Scale Update Committee. Ideally this sort of problem would be settled in a free market, but that is challenging because patients have poor information about the costs and quality of services in different healthcare settings.

Second, the actual efficacy of the procedure is nowhere to be found in the computation of payment. For the most part, therapies are deemed to be either appropriate or not as determined by whether they are medically necessary for the patient. For medications, this definition usually equates to FDA approval although there are some counter-examples. Under this system, assuming CMS has done a good job of estimating costs, there is a perverse financial incentive for healthcare providers to choose whichever medically necessary therapy costs the most since providers will make a percentage of that back in profit.

While this is the standard in the United States, there are systems in the UK and elsewhere designed to set reimbursement based on Quality Adjusted Life Years (QALY). This is an attempt to quantify the efficacy of a medical treatment and incorporate that into the reimbursement computation. It addresses the second issue associated with health insurance reimbursement decisions, but is still susceptible to bias in the estimation of efficacy.

Is it possible to take advantage of free market ideas to drive efficiency and improve health?

A free market economy becomes a force for good when the goals of earning money and benefiting society can be aligned. However, the incentive structures in healthcare are all wrong.

Hospitals are paid to do more, not make people healthier.
Physicians can earn more by choosing more expensive therapies.
Drug developers earn more from palliative drugs that must be taken continuously rather than cures.

As things stand now, market forces are often in strong opposition to the benefit of society. Momentous changes such as risk sharing arrangements between hospitals and payers, public health exchanges, "big" data and the availability of data to patients and the imposition of payment penalties based on quality metrics are all leading to changes in important financial aspects of healthcare decisions. If their combination leads to better alignment between financial incentives and patient health, then we are at the beginning of a healthcare renassaince in the United States.

Originally published on EHR Analytics, 2014.

List of procedures that cost Medicare (and patients) the most money

2014-10-16T00:00:00.000Z

Data from the recently released Medicare provider utilization and payment data contain cost information for over 5000 procedures performed for seniors in the United States in 2012. The data is a little complicated and contains some biases. However, with a little care it is possible to identify the most expensive procedures and get an understanding of the monetary value our healthcare system has assigned to human life.

The costs that are reported in the data set are for a single procedure. However, if a course of therapy requires that a procedure be repeated multiple times, then the real cost is higher. In order to get an understanding of annual cost, we need the number of procedures in a course of therapy. To get this we divide the number of times a procedure was completed by the number of unique patients receiving that procedure (both available in the data). This is still an underestimate because we only have data on procedures performed in 2012 while the course of therapy may have extended into either 2011 or 2013.

The table lists 37 therapeutic options that cost more that $10,000 for a course of therapy in 2012. Focusing just on treatments for cancer that are listed in the table (in orange) we see that the value of a month of life varies from $2,333 for Oxaliplatin to treat colorectal cancer to $29,709 for Ipilimumab to treat melanoma.

By not specifically setting a price on human life, we are allowing the free(ish) market to make those decisions. That isn't inherently good or bad, but free market in the American healthcare system is distorted by perverse incentives and high levels of information asymmetry. The patients, physicians, payers and makers of therapeutics all have vastly different levels of understanding of the value and cost of therapies. In a later article, we will look at some specific examples of how this leads to therapeutic decisions that aren't necessarily optimal for patients.

Expensive Therapeutics. Rows colored in orange are cancer medications, rows in blue are not medications. The value "continuous" in the column "boosted overall survival" means that the drug must be taken continuously for the life of the patient. Overall survival is drawn from cancer.gov and therapeutic indication is drawn from the FDA label for the medication.

Code	Procedure	Therapeutic Indication	Average Allowed Payment	patients in dataset	Bosted overall survival (months)	Cost per month of life ($)
J7192	Factor viii recombinant NOS	Hemophelia	249877.5	357	continuous	20,823
J7187	Humate-P, inj	Hemophelia	211461.2	12	continuous	17,622
J7193	Factor IX non-recombinant	Hemophelia	186561.1	13	continuous	15,547
J7195	Factor IX recombinant	Hemophelia	185380.8	43	continuous	15,448
J1786	Imuglucerase injection	Hemophelia	133782.8	35	continuous	11,149
J7190	Factor viii	Hemophelia	132205.5	55	continuous	11,017
J9228	Ipilimumab injection	Melanoma	118836.5	71	4	29,709
Q2043	Sipuleucel-T auto CD54+	Prostate cancer	63459.8	274	4.1	15,478
Q3025	IM inj interferon beta 1-a	Multiple sclerosis	28033.83	17
J1561	Gamunex/gamunex c	Primary Immuno- deficiency	26335.92	1393	Continuous	2,195
J9043	Cabazitaxel injection	Prostate Cancer	25994.92	29	2.4	10,831
J2353	Octreotide injection, depot	Cancer supportive care	20100.57	906
J0490	Belimumab injection	Lupus	19702.11	122
J2796	Romiplostim injection	Chronic ITP	18650.44	82	Continuous	1,554
J9055	Cetuximab injection	colorectal cancer head/neck cancer	18291.55	708	colorectal: 1.1 head/neck: .8	colorectal: 16,628 head/neck: 22,864
77523	Proton trmt intermediate	various cancers	17700.16	887
L8687	Implt nrostm pls gen dua rec		16881.23	95
J9305	Pemetrexed injection	lung cancer	16869.75	3048	non-squamous: 2.8 mesothelioma: 2.8	non-squamous: 6,025 mesothelioma: 6,025
0182T	Hdr elect brachytherapy	various cancers	16852.01	540
J2323	Natalizumab injection	Crohn's disease, multiple sclerosis	16348.39	2716
J1572	Flebogamma injection	Primary Immuno- deficiency	15625.81	209	Continuous	1,302
J9310	Rituximab injection	Leukemia, lymphoma	15078.95	35654
J2562	Plerixafor injection	non-Hodgkin lymphoma, multiple myeloma	14181.41	12
J9355	Trastuzumab injection	HER2 Gastric cancer HER2 Breast cancer	13762.43	2773	Gastric: 2.4	5,734
37231	Tib/per revasc stent & ather	revascular- ization surgery	13452.13	441
J1745	Infliximab injection	Crohn's, Ulcerative Colitis, Arthritis	13430.7	42645
J1568	Octagam injection	Primary Immuno- deficiency	12171.14	1397	Continuous	1,014
77600	Hyperthermia treatment	various cancers	12111.14	77
J1569	Gammagard liquid injection	Primary Immuno- deficiency	11939.5	1586	Continuous	995
J1459	Inj IVIG privigen 500 mg	Primary Immunodeficiency, Chronic ITP	11832.02	663	Continuous	986
37227	Fem/popl revasc stnt & ather	revascular- ization surgery	11535.57	4319
J9263	Oxaliplatin	Colorectal cancer	11196.69	8164	4.8	2,333
J3262	Tocilizumab injection	Rheumatoid arthritis	10997.55	2216	Continuous	916
36516	Apheresis selective	Various	10901.78	42
J9264	Paclitaxel protein bound	pancreatic cancer, lung cancer, breast cancer	10605.92	1041	pancreatic: 1.8	pancreatic: 5,892
J0129	Abatacept injection	Arthritis	10231.23	13916	Continuous	853
J2357	Omalizumab injection	asthma, idiopathic urticaria	10042.75	2770	Continuous	837

Originally published on EHR Analytics, 2014.

Medicare release of 2012 Part B data

2014-10-16T00:00:00.000Z

Recently the Centers for Medicare and Medicaid Services (CMS) released Medicare provider utilization and payment data to the public. The data released accounts for a relatively small and biased view of total Medicare spending (Table 1). Nonetheless, it does make public the cost, to both Medicare and to patients, of most of the procedures performed for seniors in the United States.

Table 1 - Understanding the data. The Medicare data release contains a subset of Medicare Part B. Spending for the “Part B data release” row was computed directly from the data. Otherwise, spending reported here was derived from this article.

Medicare	Spending (billions of dollars)	Description
Part A	$214	Hospital fees, home health, skilled nursing
Part B	$145	"Medically necessary" services and supplies
Advantage Plans	$123	Accountable care and other risk sharing arrangements
Part D	$54	Drugs
Part B data release	$77	Filtered to protect patient anonymity. Contains cases where physician performed procedure > 10 times.

We have seen a number of articles recently utilizing the data (1,2) to focus on the large sums of money being collected from Medicare by relatively few physicians. This should not, however, be particularly surprising, as the 80/20 principle is a relatively ubiquitous phenomenon in many industries. On the other side of the healthcare coin, similar observations (1,2) have been made regarding a relatively small percentage of patients who account for a large percentage of overall healthcare spending.

Before digging into this data it is worth understanding some of its weaknesses. First, the data has been filtered in order to protect the anonymity of patients. Specifically, if a physician has performed a particular procedure less than 10 times, then data on that procedure from that physician isn’t available. Second, the data covers only Medicare Part B “fee-for-service” claims from 2012.

Because of the type of data filtering being used, non-specialists – who may not meet the 10-time threshold on any particular procedure – appear to collect a lower percentage of the Medicare dollars than they actually do. This filtering plus unavailable data from Parts A, C and D make articles claiming that particular portions of the pie go to particular individuals somewhat suspect.

While in the process of analyzing the CMS physician payment data it is tempting to focus on the practices of individual physicians – the data lists the names and work addresses of over 880,000 different healthcare providers – that should not be the main purpose of this data. In addition to that, because of the biased filtering and missing data from Medicare Parts A, C and D, it isn’t even what the data is best suited for.

In a companion article to this one, we look at the prices of various treatments that are approved by the FDA and paid for by Medicare. This data release offers the opportunity to examine the way that Medicare pricing combined with healthcare market forces lead to decisions about the value of human life. If by releasing it CMS makes it possible to move our healthcare system closer to a free market – in which providers compete on prices and quality metrics that are transparent – then we are closer to addressing the inefficiencies and huge costs of our healthcare system.

Originally published on EHR Analytics, 2014.

A rapidly changing landscape is leading to uncertainty and opportunities throughout healthcare

2014-04-24T00:00:00.000Z

Huge volumes of data about patient health from electronic medical records (EMR), high-throughput molecular data, insurance claims, the “quantified self” movement, and social media, are rapidly becoming available. At the same time, changes in financial incentives such as the utilization of healthcare exchanges, the creation of ACOs (Accountable Care Organizations) and the growth of clinical research networks are driving changes in business models that will have far reaching consequences. Currently there is a gap between the huge quantities of health data and the discovery/validation of new approaches to managing the health of patients and patient populations. There is a tremendous opportunity to develop new statistical methodologies to pull information out of the data that can be used to improve the efficiency and effectiveness of healthcare delivery.

Quality improvement by hospital systems. One of the challenges facing physicians today is deciding which “standard of care” to follow. In many cases there are numerous therapeutic options for a patient, all of which are acceptable. Published studies addressing the question are often sparse, so the decisions are commonly made based on marketing materials provided by the pharmaceutical companies themselves. In addition, in a “fee-for-service” environment, there is a perverse financial incentive to choose the most expensive therapeutic. However, for hospital systems that accept some of the expense when patients do not respond well to treatment, such as ACOs, incentives are quite different. Even for traditional “fee-for-service” institutions, new federal regulations and “meaningful use” criteria are driving a need to identify and impose optimal care. How should “optimal care” be defined? How do health systems utilize patients’ health records to identify treatment decisions that lead to optimal care? How can healthcare systems design trials, run from the EMR or other automated data sources, to prove or disprove the hypotheses generated from retrospective analyses?

Example. Our modern healthcare system is fragmented. This leads to different providers following different patient outcomes that are tied to the diseases for which they are responsible. A cardiologist may prescribe a statin for high cholesterol, but if the patient taking that statin gets muscle aches they are more likely to go to their family practitioner; the physician who originally prescribed the medication might never even find out about the side effects! If there is institutional motivation, the health record can be used to track and measure overall health. The proxy for “overall health” in this scenario may very well be defined as lower utilization of hospital resources; In a perfect world, patients will agree that this is a good proxy.

Recruitment for clinical studies. Typical large trials are run at many different clinical sites in order to ensure the accrual of enough patients for the study. In this setting there are often numerous sites that fail to recruit even a single patient. The availability of electronic health records creates the opportunity to directly identify the right patients for a new trial and to target recruitment efforts. This can simultaneously cut down on trial startup expenses and boost recruitment rates. Networks of hospital systems are already building this capability and will have tremendous advantages when competing to run certain types of clinical studies. However, electronic health records are inherently messy and incomplete. What is the best way to cut through the noise and identify the right patients? How early in the course of disease can patient populations be identified?

Example. PCORnet is a group of hospital systems who have obtained federal funding to develop an automated system for pooling and sharing the health data of individual patients. It is designed to automate many of the steps involved in conducting clinical trials. If you are a fan of NPR, Diane Rehm devoted a show to this concept (and PCORnet specifically); you can listen to it here.

A separate, innovative approach to patient recruitment has been developed through the participation of the patients themselves. Last year a social networking web site, Patients Like Me, and a clinical research organization, inVentive Health, formed a partnership to advertise recruitment for clinical trials directly to the patients.

Preventive medicine. A systematic approach to preventive care will be important for those healthcare systems who are trying to minimize the future disease burdens of their patient populations. Historical health data, high-throughput molecular data, information from social media, data from “quantified self” devices, and even purchasing data from credit cards can all offer insight into the current and future health of patients. *Which patients within the health system are most susceptible to future disease? What sources of data are best able to identify those patients? What interventions are best able to prevent bad outcomes in the long term?*Integrating all of the relevant sources of information – and filtering out the irrelevant sources – in order to build disease specific models of risk will be critical to identifying patients who are appropriate for preventive medicine efforts.

Example. Consider the announcement from CVS that they will stop selling cigarettes in order to better position themselves as a healthcare delivery company. As they begin to provide healthcare services they will accrue health data on their customers which can presumably – barring legal restrictions – be tied to other purchases. Purchases of candy bars, shampoo and razors can easily become part of your electronic health record. If one of the first signs of dementia is neglect of personal hygene, CVS may be the first to know when grandma is developing Alzheimer’s disease! CVS is not alone in this new business model; Walmart, Target and Walgreens all have clinics in at least a subset of their stores.

Precision medicine. Until now, clinical research has favored a “one size fits all” approach to the development of novel therapeutics. This is driven by a desire to maximize the market share of any new drug; if the drug can only be given to the patient sub-population who passes a companion diagnostic test, then the drug has a smaller market. However, the cost of development is increasing exponentially and the chance of eventual FDA approval is dropping. Acceptance of a smaller market share in trade for an improved chance of FDA approval (and possibly higher market penetration) is driving an increasing willingness in the pharmaceutical industry to develop drugs with companion diagnostics. Companion diagnostics are often based on high-throughput molecular data such as DNA mutation, RNA expression, metabolomics and proteomics. What is the best way to integrate high-throughput molecular data with clinical data to ensure the identification of the optimal subpopulation for a new therapeutic? Can we make the case for a new therapeutic within the context of the new financial and regulatory incentives faced by healthcare systems?

Example. The FDA lists 9 different drugs and 19 different drug – companion diagnostic combinations that are approved. However, they list 154 drug – gene pairs for which particular versions of the gene lead to potential adverse events. Some of these are serious events. For example, some people have a variant in a gene called CYP2D6 that causes Codeine to be metabolized into morphine very quickly. In children, that process can lead to lethal doses. Unfortunately, identifying genetic variants that lead to serious adverse events does not automatically lead to the requirement that the gene be tested before the drug is given. It will be up to providers to decide what is best for their patients and payers to decide which tests will be reimbursed.

I have discussed only a few places where the combination of federal regulation, changing incentives and “big” data are coming together to transform healthcare as an industry. However, when combined these constitute large shifts in business models with the potential to leave companies who stick to old approaches in the dust. It is impossible to know where healthcare in America is going, but it is clearly going somewhere.

Originally published on EHR Analytics, 2014.

Publicly Available Electronic Health Data

2014-04-18T00:00:00.000Z

This entry in the blog is a list of electronic health data sets that are available, in some way or another. Some are freely available, some require fees and some require special connections.

Data source web site. There is an Interactive Compendium of Health Datasets for Economists web site maintained by the University of Oxford that should be mentioned in this context. It provides links to a number of health related datasets for the purposes of health economics research. There is a nice search feature that allows filtering of the known data sets based on a number of different fields. For example, I found one data source containing longitudinal primary care data at the level of the individual.

Free, publicly available

· A recent release of health data by the US Centers for Medicare and Medicaid Services made a large splash in the mainstream media. That data does not give patient level records, but it does represent very granular information about providers. The data is split into three groups: Physician and Other Supplier, Inpatient, Outpatient.

· The PhysioNet challenge is an annual competition focused on computer analysis in the field of cardiology. It has been running since 2000, and a few of the competitions have involved electronic medical records.

· The Heritage Provider Network released some insurance claims data as part of a competition to predict which patients will be admitted to the hospital within the next year. Claims data is what the hospitals report to insurance companies and is utilized almost exclusively for billing. Some studies have suggested that it is inferior in some ways for the purpose of identifying and tracking patient disease. There are certainly strong financial incentives for hospitals to distort the picture presented in claims data as long as they avoid fraud.

· The Pooled Resource Open-Access ALS Clinical Trials (PRO-ACT) database is a collection of data from studies of Amyotrophic lateral sclerosis. Generally, clinical trials data is more extensive, more complete and more accurate than typical electronic medical records data. However, there is a lot of oversight of patients who are on trials, and patients have to volunteer to join the trial. This means that there are differences in the likelihood that patients on trials will stop taking their drugs as well as more general demographic differences between patients on trials and the general patient population.

· The Agency for Healthcare Research and Quality (AHRQ) has made available a number of data sources associated with its Healthcare Cost and Utilization Project. These data sources include limited information about a large collection of hospital discharges.

· Every year I2B2 hosts a competition designed around natural language processing of electronic health records. This year there are two challenges. One focused on de-identification and another focused on identifying risk factors for heart disease. You need to register before the contest begins in order to get access to the data, and you have to agree to the contest rules.

Connections required

· If you have or can find a research collaborator in Canada, the Canadian Institute for Health Information makes available most of the hospitalization data from Canadian hospitals.

Fees required

· A plan to share the British national health data broadly has been put on temporary hold. However, the British National Institute for Health Research does make at least some of the British health system data available under the name Clinical Practice Research Datalink. I am told that fees for access to this data are around $100K/year, but I could not find pricing information online.

· I examined the New Zealand National Minimum Dataset in a previous article. I have since found out that it is available for a fee that is determined based on the hours required to pull the data (priced at around $70/hour).

If I find out about any more, I will post them.

Originally published on EHR Analytics, 2014.

EMR Data access: The New Zealand National Minimum Dataset

2014-03-18T00:00:00.000Z

Before I continue I want to acknowledge that, for good reason, it is difficult to get access to electronic medical records data. That makes it very tough for statisticians who are interested in working on data analysis problems in medicine. For those of you playing the home game, I will try to use at least moderately accessible data wherever possible, though there will be times when that is impossible. Until something more accessible comes along, I plan to use the New Zealand National Minimum Dataset. This dataset comes with a number of challenges, but at least it is accessible (for a fee). The version I have contains data from patient encounters at New Zealand hospitals occurring in the years 2006 through 2012.

Positive features of this dataset:

You can get it.
New Zealand is an island, so we have the full account of admissions for most of the population.
The data set is moderately large - 3.55 million visits from 1.3 million patients.

Negative features of this dataset:

We do not have information about patients visits that occur outside of the hospital setting.
Lab values, medications, physicians' notes and many other interesting features of medical records are not available.
The data have been filtered.
We only know when someone has died if they died in the hospital

There are only a few other options for analysis of medical records if you are among the majority who do not to have access at your job. There are periodic contests in which small medical records data sets are published along with a very specific task. These include a Kaggle contest in which contestants are challenged to identify patients with diabetes, and annual contests published by i2b2. This year's i2b2 challenge is split into two tracks, one to de-identify electronic health data and another to identify risk factors of heart disease over time. A final data set that is worth mentioning even though it is not officially available is from the British National Health Service (NHS). They are planning to make available all electronic health data from NHS patients. If you are interested, keep an eye on care.data website for details and access as they emerge.

I am lucky to have access to some medical records in my job and have therefore not made a careful study of the publically available EHR datasets. Readers are encouraged to post information about other publically available record sets (even if they require a fee for access) in the comments section.

New Zealand National Minimum Dataset.

There are some features of the New Zealand data set that are important for understanding what can be done. All ages recorded in the dataset are between 18 and 65. This will impose some limits on our ability to look at diseases that are prevalent in children or in old age. In addition, the gender ratios follow an unusual pattern (Figure 1). There are vastly more women than men in the dataset between the ages of 18 and 40. This produces some odd results when looking for relationships between diseases, as we will see in the next article.

Figure 1. Age / gender distribution. There are vastly more visits by women between the ages of 18 and 40 than there are men.

Imperfections. This dataset has been very carefully cleaned. For example, there is not even one missing age value and there are only 9 patient visits associated with unknown gender. This is quite unusual in medical records and suggests that records with missing data are simply not being reported. Keep in mind that we are working with a lot of data; if you had to manually enter the gender of 1.3 million people a total of 3.5 million times, how many missing values would there be? How many mistakes? To get an idea of how accurate these data are, we can look for internal discrepancies.

Consider, for example that a patient’s gender is entered every time they are admitted to the hospital. We can therefore look for instances in which the gender of a patient changes from one visit to the next. Of the 400,000 patients with two or more visits, there are 400 patients who have more than one gender recorded in the record. Admittedly, this would be accurate for patients who have undergone a sex change, but among the 400 are 114 whose gender has changed 2 or more times. Surely that level of indecision about one’s sexuality is exceedingly rare! There are also 2 men in the record who have been admitted to the hospital to deliver a baby.

In addition to gender, we can look for discrepancies in age. The record contains separate entries for age at discharge (in years) and date of discharge. These two entries can be used to estimate a patient’s birth date to within 1 year. There are 2,152 patients whose estimated birth dates lead to impossibilities – the lowest and highest estimates vary by more than one year. There are 109 patients with estimated birth dates separated by more than 10 years!

I am pointing out flaws in the record not to disparage this particular record set – this level of accuracy is on par with other record sets that I have looked at. The main point is that, no matter what is done with electronic medical records, it must be done with the understanding that there are errors in the data. An algorithm or model that depends absolutely on any particular feature of the data is guaranteed to make mistakes. How critical those mistakes are will depend heavily on the application.

Figure 2 Readmissions versus days of use. There are generally two different ways for a patient to become “expensive” from the point of view of a hospital system. If they are often readmitted within 30 days of discharge or if they have diseases that demand very long hospital stays.

Expensive patients. Based on a previous article, we know that hospitals are very keenly interested in avoiding early readmission to the hospital. Can we identify those patients who have the highest rates of hospital admission? One of the shocking features of healthcare is the amount of resources that can be utilized by the sickest patients. The largest number of independent visits recorded for any single patient in the record set is 1175. Since we are looking at seven years of data, this implies that there are individual patients who are readmitted to some hospital almost every other day!

There is an often cited “80/20 rule” (attached to Pareto distributions) which states that in many real world situations 80% of the resources are controlled by 20% of the population. This rule has been specifically applied to healthcare expenses in political debates and there is evidence that it is accurate. However, in the New Zealand dataset we find that the 20% of patients who had the most hospital admissions accounted for only around 53% of total. This is perhaps due to the limited number of days available in the calendar. Another possibility is that admission to the hospital is not a good proxy for healthcare expenses.

Even though it is somewhat at odds with the Centers for Medicare and Medicaid Services rule about 30 day readmissions, it seems likely that total time spent in the hospital is a better proxy for total expenditure. We have both start and end dates for all admissions in the data set, so we can compute the total amount of time spent in the hospital for each patient. You can see along the bottom of Figure 2 that there is an obvious group of patients who have very few admissions, but who spend extraordinary amounts of time in the hospital. Going back to the 80/20 rule, we find that 71% of the resources (hospital days) are utilized by the top 20% of patients. This is still not quite the level reported in American media, but it is within range. It is possible that the average cost per day of patients who are in the hospital is higher for patients who are there more; this might explain the discrepancy. It even seems likely since those patients are probably sicker. It is also possible that some feature of New Zealand’s medical system leads to more egalitarian utilization patterns.

Based on what we’ve seen, we can identify patients who have been expensive in the interval from 2006 through 2012. There are a host of questions that come out of this analysis, but very few answers. Are those same patients going to be expensive in 2013? What are the features of those patients? Are they associated with particular diseases? Can we tell which patients are on a path to spend a lot of time in the hospital? If so, are there preventive medicine options that can be implemented to head that off?

With the new rules regarding preexisting conditions, insurance companies are no longer allowed to use previous expenses to set premiums. However, some hospitals are taking on part of the expense of caring for patients. What will they do to decrease the cost of these patients? Ideally they will seek out more efficient and successful ways to care for them. Hopefully the availability of data and the importance of the question will lead to solutions that work for everyone involved.

Originally published on EHR Analytics, 2014.

Disruptive Transformation for Hospital Systems and a Couple of Places where Statisticians can Help

2014-03-06T00:00:00.000Z

For many years hospitals and physicians have been paid by insurance companies for each procedure they perform. This may seem reasonable since every procedure performed, from discussing a patient's disease to performing the most complicated surgery, requires resources. However, it creates a perverse set of financial incentives for physicians and hospital systems. Sicker patients lead to more procedures which lead to more revenue. The financial incentive is to make patients sicker!

Physicians and hospital administrators recognize the wrongness of this incentive structure. Only criminally anti-social individuals would actively pursue "upselling" as a legitimate means of increasing hospital revenue. Therefore, in order to obscure and minimize the effect of this financial incentive, physicians are shielded from the costs of the procedures they perform and hospital administrations typically do not monitor the health of their patient population.

There is a movement in healthcare to impose financial incentives for healthcare providers to make patients healthier. Recent changes in (1) the rules by which the Centers for Medicare and Medicaid Services (CMS) must operate and (2) federal law regarding the implementation of electronic health records, are beginning to make this change a reality.

As of 2012 CMS can work with hospitals or groups of physicians to create "Accountable Care Organizations" (ACOs). Under this payment structure, care providers are given a fixed fee for each patient for whom they are responsible; if they can save money in the care of that patient they get to pocket the savings. This is similar to the fee structure of Health Maintenance Organizations that were reviled by patients in the 1980's because they were financially incentivized to minimize patient interactions.

There are some key differences which, if taken advantage of, can lead to a different outcome for ACOs. First, with ACOs and other "risk bearing" healthcare organizations there are penalties when patients do not do well. This leads to a question into which statistics can offer insight; since everybody is different, what does it mean for a patient to be doing well? Second, the opportunities for communication between hospital systems and patients have vastly improved since the 1980’s. Try searching for “frustrated with hospital” on twitter and you will readily see that communication from the patient to the hospital is already very robust. Third, due to the Affordable Care Act, there is a growing percentage of the population who are directly responsible for choosing their own health insurance. It is natural to demand the most expensive insurance from one’s employer if a choice of health insurance provider is not part of the hiring process. However, when an individual is deciding between plans with vastly different prices, choosing a plan that encourages maximizing the number of procedures no longer seems like the obvious choice – it shouldn’t have been anyway.

In addition to changing fee structures, as of 2012 healthcare organizations are required to maintain electronic medical records. The original intention of this law was to encourage the free exchange of health information between providers in order to minimize duplication of effort; if a patient has an x-ray at one hospital, it should not be repeated the next day if they show up at a different hospital. In practice, this objective has not yet been realized because every hospital has its own EMR and those systems are not interoperable – even if they were purchased from the same vendor. However, what has been created is a vast trove of data about the health of individual patients. The potential of this huge amount of data to affect all aspects of healthcare cannot be overstated. In particular, risk bearing hospital systems now have both the financial incentives and the necessary data to track the health of their patient population and be proactive in the treatment of disease. Successful physician-statistician collaborations are needed to turn this data into information that hospital systems can act upon.

Driving uptake of IT and Statistics in Healthcare

The strongest and earliest driver encouraging hospitals to implement electronic medical records originates from CMS in the form of a couple of different penalties. The “meaningful use” requirement is being implemented in three increasingly strict phases (phase 1, phase 2 and phase 3). Those hospitals that are deemed not to be utilizing their electronic medical records in a meaningful way will be penalized, beginning in 2015, with a 1% decrease in CMS payouts. The penalty increases by 1% yearly up to a total of 5% for consistent failure to achieve meaningful use – tens of millions of dollars for hospitals with large Medicare and Medicaid populations. Most of the meaningful use definitions require solutions that are straightforward even if they are technically complicated to implement. As of early 2014, I am not aware of any “meaningful use” applications that involve statistical solutions, though I can imagine improvements to the current versions that might. Here I discuss a statistical approach to identifying homogeneous groups of patients (one of the elements of meaningful use in phase 2). Whether improvements like these are financially viable will depend heavily on the way that financial incentives are structured for hospitals.

The second penalty (again a 1% incrementally increasing penalty), and the one that has led the industry to seek out statistical solutions, is a reduction in payment for hospitals with too many patients who are readmitted within 30 days of discharge. There has been an explosion of statistical models attempting to predict early readmission – there is an open access survey in JAMA for those who want greater detail. To my knowledge, all of the models attempting to accomplish this are logistic regressions which, in their final form, rely on a clearly defined set of data (independent variables) with which to make predictions. Implementing these models in practice is challenging because electronic records do not follow fixed standards, patient populations vary significantly between hospitals, and every hospital record system is plagued by missing and incorrectly coded data. Finally, it is not always clear how far back into a record one must go. A typical statistical approach to modeling the time varying state of a patient is to assume that all the relevant information for predicting the future is available in the present (see hidden Markov model and memorylessness). However, if two patients come to the hospital with skin infections, and one was diagnosed years earlier with diabetes, the severity of their infection and their chances of returning within 30 days are very different.

I have described some of the disruptive changes that hospitals are undergoing as a result of changing incentives and the availability of electronic health records. The availability of this data will disrupt healthcare delivery at all levels; insurance companies, contract research organizations, pharma, regulators and consumers are all seeing (and will continue to see) disruption.

Perhaps the most exciting thing for statisticians is the availability of a vast array of statistical challenges in the healthcare industry that are financially viable, able to tolerate uncertainty and just downright fun to work on. We will likely never get to a point where computers can be trusted to make medical decisions for patients, but a tricorder reminiscent of Star Trek might be just around the corner, and even a 1% increase in efficiency for the half-trillion dollar drug discovery industry would be tremendously valuable.

Originally published on EHR Analytics, 2014.