Unbiased Analysis of Today's Healthcare Issues

Falsification Test for Instrumental Variables

Written By: Jason Shafrin - Sep• 09•15

Should instrumental variables (IV) be used for real-world evaluation of the comparative effectiveness of different studies?  It depends on who you ask.Garabedian et al. (2014) state

Although no observational method can completely eliminate confounding, we recommend against treating instrumental variable analysis as a solution to the inherent biases in observational CER studies.

On the other hand, Glymour, Tchetgen, and Robins (2012) state:

Given that it will often benearly free to conduct IVanalyses with secondary data, they may prove extremely valuable in many research areas . . . [however if IV] is uncritically adopted into the epidemiologic toolbox, without aggressive evaluations of the validity of the design in each case, it may generate a host of false or misleading findings.”

IV approaches are problematic if you have a weak instrument (i.e., it is weakly correlated with sorting into treatment) or if the exclusion restriction (i.e., that the IV is uncorrelated with outcomes except through the probability of receiving treatment) does not hold.  One can readily test the strength of the relationship between the instrument and the probability of treatment assignment (see Stock and Yogo).  Testing the exclusion restriction, however, is harder to test.

A paper by Pizer et al. (2015) recommends using falsification test to see whether it is likely that the exclusion restriction holds.  He writes:

Although falsification tests in general can take many forms, there are two particularly useful strategies for testing the exclusion restriction in IV CER studies: (1) investigating an alternative outcome that ought not to be affected by the treatment under study but would be affected by potential confounders that might be correlated with the proposed IV; and (2) investigating an alternative population that again ought not to be affected by the treatment but would be affected by potential confounders.

One example he gives is the US of IV to measure stroke outcomes among patients using atrial fibrillation.

Garabedian et al. (2014) argued that practice pattern IV studies are often vulnerable to bias because they fail to control for one or more of the following patient characteristics: race, education, income, age, insurance status, health status, and health behaviors. For example, if health behaviors are correlated with anticoagulant prescribing patterns and the outcomes under study, this could indeed be a problem. However, patients without atrial fibrillation but who have carotid artery disease are also at elevated risk for stroke and should not be treated with anticoagulants. If anticoagulant prescribing patterns are unrelated to stroke outcomes for carotid disease patients, then it is less likely that confounding health behaviors are correlated with anticoagulant prescribing patterns (panel B of Figure 2). Instead of using an alternative population (those with carotid disease), another option would be to choose an alternative outcome that should not be affected by the treatment but would be affected by health behaviors (e.g., incident lung cancer).

Not that although it is never possible to prove for certain that there is no confounding, falsification tests provide evidence that the IV exclusion restriction is valid.


Does increased use of prescription drugs lower medical costs?

Written By: Jason Shafrin - Sep• 08•15

There is a belief that providing better care can reduce cost, at least somewhat.  For instance, some claim that better primary care can avoid unnecessary hospitalizations.  But can increased use of prescription drugs lead to decreased medical spending?  This is exactly what a paper by Roebuck et al. (2015) find.  They write:

We found that a 1 percent increase in overall prescription drug use was associated with decreases in total nondrug Medicaid costs by 0.108 percent for blind or disabled adults, 0.167 percent for other adults, and 0.041 percent for children. Reductions in combined inpatient and outpatient spending from increased drug utilization in Medicaid were similar to an estimate for Medicare by the Congressional Budget Office.

Prescription drugs are expensive; nevertheless, in aggregate they are likely to reduce non-drug medical spending and improve health.


  • Christopher Roebuck, J. Samantha Dougherty, Robert Kaestner, and Laura M. Miller. Increased Use Of Prescription Drugs Reduces Medical Costs In Medicaid Populations. 10.1377/hlthaff.2015.0335 HEALTH AFFAIRS 34, NO. 9 (2015): 1586–1593

Selection on Moral Hazard

Written By: Jason Shafrin - Sep• 07•15

The terms adverse selection and moral hazard are well known within the field of health economics.  But what is “selection on moral hazard”?  Amy Finkelstein explains using the following analogy:

In the context of an all-you-can eat restaurant, traditional selection is that people with big appetites are more likely to go to all-you-can-eat restaurants.  Selection on moral hazard is the idea that even if you have an average appetite, you know that when food is free on the margin–the marginal price of an additional entree is zero at an all-you-can-eat restaurant–you’re going to consume a lot more than you usually do when things are priced a la carte.  So people who tend to eat a lot tend to eat a lot more when the price of food is lower also find an all-you-can-eat restaurant appealing.  Selection on moral hazard is thus selection on the slope, or on the price sensitivity of demand, ratherthan “traditional” selection on the intercept, or the level of demand.

But does this type of thing happen in real life.  Finkelstein and co-authors (2013) use data from Alcoa’s employee’s health insurance options, choices and medical claims and found that individuals who were more likely to increase their use of health care services when the price is low are also the ones who are most likely to choose plans with lower cost-sharing. In fact, Finkelstein claims that selection on moral hazard “…was almost as important as traditional adverse selection” in plan selection.


Convenience improves adherence

Written By: Jason Shafrin - Sep• 03•15

This is one of the goals of Appointment-based medication synchronization (ABMS).  These systems vary in their implementation but they have three common features:

  • All medications refills come due on the same day of the month.
  • Pharmacies place regular call to remind the patient to fill their prescription, typically 5 to 7 days before the scheduled pharmacy visit
  • Patients schedule appointments to pick up medications (or have them delivered). Medication therapy managment and/or disease management programs can be integrated into this appointment.

The question is, does it work?  Using retrospective data from Ohio-based drugs stores, a study by Holdford and Saxena found striking results:

Mean PDC scores ranged from 0.73 to 0.91 for ABMS patients (n=205 to 716) and from 0.57 to 0.71 for usual care depending on the medication class. The percentage of adherent individuals (i.e., PDC≥0.80) was 55% to 84% for ABMS participants and 37% to 62% for usual care. Odds of adherence was 2.3 to 3.6 times greater with ABMS. Usual care patients became nonpersistent (61% to 74%) more often than ABMS patients (33% to 44%) with hazard ratios of nonpersistence being 0.39 to 0.67 for individuals in the program.

For patients with multiple chronic conditions, simplifying their medication regimens is a laudable goal.  Future research should examine whether these improvements in adherence are also correlated with improved health outcomes and lower costs of medical care.



A super EMR?

Written By: Jason Shafrin - Sep• 02•15

Sources of health care data are proliferating.  The previous standard–medical charts–are being augmented with information from digital sensors, patient reported outcomes, and genetic information. Wouldn’t it be nice to have all that information in one place?  That is what Salesforce is thinking.  Fortune reports:

Salesforce announced on Wednesday a new patient relationship management platform that it calls Salesforce Health Cloud…Salesforce Health Cloud combines data from multiple sources—electronic medical records, medical devices, even wearables—into a single location. The idea? By having all the information in one place, health workers will have a more complete view of the patient and, in turn, be able to make smarter care decisions, intervene earlier if issues arise, and collect data along the way for effective treatments. The software also puts that information in the hands of the patient through mobile applications.

There are a number of unanswered questions however.  For instance, is it secure? Putting all your data in one place makes it an ideal target for hackers.

Who has access to the Health Cloud?  The patient and his physician clearly; but in the era of patient-centered medical homes, the patient may be treated by tens of doctors and numerous nurses and medical assistants.  Will all these individuals have access to the cloud?  Who will be able to upload data and delete it?  Will the Health Cloud comply with the numerous federal regulations and help physicians meet meaningful use requirements?

A centralized data repository clearly has the potential to produce significant social value.  How the the Health Cloud is implemented–however–will ultimately determine whether this is a significant advance or a pie in the sky dream.


Mid-week Links

Written By: Jason Shafrin - Sep• 01•15

Medicare spending surges again

Written By: Jason Shafrin - Aug• 31•15

The August update to the Congressional Budget Office’s 10-year economic outlook is fairly rosy.  The deficit will ‘only’ be $426 billion, which is $59 billion less than the deficit last year and would represent 2.4% of GDP, the smallest deficit as a share of GDP since 2007.  Nevertheless, CBO still products overall US debt to rise as a share of GDP by 2025.

In health care, the news is more worrisome, even in the short-run.

In 2015, spending for Medicare (net of premiums and other offsetting receipts) will rise by $35 billion, or about 7 percent, CBO expects—the fastest rate of growth recorded for the program since 2009 (after adjustments are made for shifts in the timing of certain payments). Part of that increase reflects the fact that certain provisions of the ACA that reduced the rate of growth in Medicare spending have been implemented already. Those provisions will continue to constrain Medicare spending, but to roughly the same extent each year, so they are no longer reducing its growth rate. In addition, the increase in 2015 reflects growth in the number or cost of services furnished to Medicare beneficiaries, although data are not yet available to show how much of that growth is attributable to changes in hospital admissions, physician visits, prescriptions of expensive new drugs, or other health care services.

The Medicare Access and CHIP Reauthorization Act of 2015 (P.L. 114-10) raised the rates Medicare pays to physicians and led the CBO to increase its projections of outlays by $159 billion for the 2016–2025 period.  However, this change is mostly just accounting tricks.  PL 114-10 ended the annual need for a “doc fix”; CBO’s alternative fiscal scenario already had assumed that the “doc fix” was a political reality.

In the long run, things aren’t much better with respect to Medicare spending due to an aging population:

Outlays for Medicare (adjusted for shifts in the timing of certain payments) remain near 3.0 percent of GDP through 2018 and then increase each year through 2025, when they total 3.7 percent.


The Cons of Restrictive Prior Authorization Policies

Written By: Jason Shafrin - Aug• 30•15

Dr Dana Goldman, a USC professor and partner at my employer–Precision Health Economics–explains how restrictive prior authorization policies can adversely affect the care patients with schizophrenia receive

What is Comprehensive Care for Joint Replacement?

Written By: Jason Shafrin - Aug• 27•15

Bundled Payments for Care Improvement (BPCI)A helpful post from Steven A. Farmer, Meaghan George and Mark B. McClellan explains.  Comprehensive Care for Joint Replacement (CCJR) is a bundled payment structure for hip and knee replacements.  CMS notes that:

2013, there were more than 400,000 inpatient primary procedures in Medicare, costing more than $7 billion for hospitalization alone.

The CCJR program creates lower extremity joint replacements (LEJR) episodes for hospital admissions for joint replacements (MS-DRG 469 and 470) that includes all payments during the 90 days following surgery.  All Part A and B services the patient receives in that 90 day period are included in the bundle.

The CCJR program is similar to CMS’s other bundled payment programs such as aptly named Bundled Payments for Care Improvement (BPCI).  However, the authors note a number of key differences between CCJR and BPCI.

  • Mandatory participation. All hospitals (with limited exceptions) in selected geographic areas are required to participate. This design enables evaluation of the program in a much broader range of hospitals than agreed to participate in the BPCI and avoids selection bias. No alternative payment model has yet been applied to an entire class of providers, and CMS intends to pursue a robust evaluation of the program.
  • Initiation occurs at hospitalization.  Whereas CCJR must begin with an inpatient admission, BPCI episodes could be inpatient, outpatient or post-acute care.
  • Providers have fewer choices with the CCJR episodes. The CCJR includes all Medicare Part A and B services, while some of BPCI models do not. BPCI offers participations a choice of episode durations (30-, 60-, or 90- day), while CCJR can only be 90 days.

One key issue with all bundled payments is innovation.  Under bundled payment–either CCJR or BPCI–provider have an incentive to adopt new technologies that lower cost and improve or do not change quality as well as technologies that do not affect cost, but improve quality.  Innovations that improve patient care but increase cost, however, will become increasingly difficult for providers to adopt if rates for CCJR episodes are fixed.  Further services that are not currently captured in billing–such as telemedicine or digital medicine technologies–would not be included in the estimated payment bundle until years in the future.

Bundled payments will incentivize providers to improve efficiency and can save Medicare money, but it risks stifling innovation and potentially harming patient care.



What is the cancer incidence rate in your state?

Written By: Jason Shafrin - Aug• 26•15

Find out at the CDC’s website. They have incidence information by cancer type and gender for all states between 1999 and 2012. Below is a sample chart you can produce with these data.