Sunday, December 13, 2015

The Beginnings Of True Pharmacosurveillance

From:  Morbidity and Mortality Weekly Report (MMWR)  October 16, 2015/ 64 (9): 1-14.
I am an advocate of real pharmacosurveillance of anyone prescribing medications in the United States.  To some that might seem like a controversial statement and it needs to be argued at a couple of levels.  There will be some privacy advocates who suggest that medication information is protected information.  Sensitive medical information can be extrapolated from prescriptions and all medical diagnoses are not seen as equal in the eyes of employers or average citizens.  In this era, the knowledge that a person has a significant amount of specific medication in their possession may also put them at risk for theft or exploitation.  There are very good reasons for making sure that this information is not leaked to the general public.  The overriding argument is public safety defined as making sure that a person seeking medical help is going to see competent physicians or in this era competent prescribers.  There are many reasons why a prescriber may not be competent and may put patients at risk, but one of the most significant reasons is that they have developed a practice or prescribing that makes them an outlier.  They are prescribing medications or combinations of medications in a manner that is not like the majority of practitioners. That prescribing pattern may be frivolous or unsafe.  In the case of unsafe patterns, the practitioner should receive immediate feedback and where necessary intervention.

Some reviews currently happen at some level in the US.  In hospitals and care systems where there is routine review of physicians, some cases are reviewed prescribing patterns are observed and they are given feedback.  That process is limited by a lack of standardization and objectivity.  Just a few cases may be reviewed when today's information technology (IT) capability allow for reviewing all of a physician's caseload all of the time.  The review is often part of a larger process like an annual review where there may be conflicting agendas like spinning the review to make the person look as good or as bad an an administrator wants them to look.  Physicians can also be contacted by managed care organization (MCOs) or pharmaceutical benefit managers (PBMs) with letters expressing various concerns.  Examples might be patients who have filled only one prescription for antidepressants, patients seeing multiple prescribers, and polypharmacy.  These letters are often poorly thought out, probably don't apply to the physician or patient at the time the letter is sent, and seem to be heavier on public relations than the technical details of prescribing medications.  In some cases these reviews can be totally inappropriate.  To cite an example, a reviewer notifies nursing staff that a patient on lithium needs follow up and immediate blood tests upon discharge because the inpatient physician has not ordered the appropriate tests when they do not have the most recent records, have not spoken to that physician, and don't know that all of the testing has been done.

Another very relevant question for pharmacosurveillance is: "Who owns the data?"  Any managed care company that I am aware of treats patient data as their own proprietary data whether they know what to do with it or not.  I gave the previous example on this blog of asking a managed care executive for permission to use deidentified brain images for teaching purposes and being told: "Why would we want you to use our data?" despite the defined teaching purpose of the institution and a long history in medicine of teaching all of the available abnormal findings for the purpose of developing better diagnosticians.  Prescription and pharmacy data has an even more clandestine history.  Most physicians were not aware until very recently that all of their prescription data was collected from pharmacies everywhere by a company called IMS America and that information was used primarily by pharmaceutical company sales forces to monitor the products being prescribed and whether their detailing people were having an impact on those prescriptions.  The individual physician was not able to see these records or look at the trends in their prescribing data over time.  The data collection was centralized only for the purpose of selling the collected data to pharmaceutical companies or (for the past 15 years or so) buried in clinic or hospital electronic health records.  The best a physician could do would be to request prescribing data on a particular patient from their pharmacy.  That might result in 20 or 30 faxed pages of lines and lines of prescriptions, usually encompassing only the most recent years.

The state may not only claim the data, but set in place mandatory rules about how practitioners collect the data and transmit it to them.  In the state of Minnesota, all practitioners treating depression are mandated to have their patient complete PHQ-9 rating scales for depression and have those results sent to the state.  The state also monitors prescribing data on stimulant prescriptions for children and mandates that any person taking an antipsychotic medication needs to sign a written consent form.   Currently 49 of 50 states participate in Prescription Drug Monitoring Programs (PDMPs) to track drugs classified on the Controlled Substances Act Schedules II - IV.  The PDMP programs were originally set up to help law enforcement identify illegal activities with controlled substances but developed into a resource for physicians who wanted to know if their patient was getting multiple prescriptions or prescriptions that increased the risk of medications that they might be prescribing.  These pre-existing systems led the CDC and the FDA to develop the Prescription Behavior Surveillance System (PBSS) to look at the trends in the prescriptions of controlled substances.   Typically all pharmacies within a state submit data on controlled substances within a week of the prescription being filled.  The PBSS categorizes all of the data into three categories: benzodiazepines, stimulants, and opioid analgesics.  Buprenorphine is classified in the opioid analgesic category even though the primary use is for treating opioid dependent patients.  Tramadol was not included in the database until it was reclassified in 2014.  There is also a miscellaneous category that includes zolpidem and carisoprodol.  I think it probably makes sense to include GABA enhancing sedative hypnotics like zolpidem, zaleplon, and eszopiclone in the benzodiazepine category.

The preliminary data from the PBSS system that was just released in October is very interesting.  In terms of a representative sample, the eight states reported cover 1/4 of the US population and represent all 4 US Census Bureau Regions.  Prescription rates were calculated as the prescription rate per 1,000 residents as given by the most recent Census data.  Perhaps most surprising is that the rate of opioid prescribing approaches nearly one prescription for every state resident in some states in some states and the rate of opioid prescribing is twice as high as the rate of benzodiazepine or stimulant prescribing.  There are two to four fold differences in state-to-state prescribing of controlled substances across the board.   The top 1% of opioid prescribers in Delaware wrote for one out of four opioid prescriptions in the state.   The top decile of prescribers account for 50-60% of opioid prescribing but that decile does not solely account for state-to-state differences.  Specialists in pain clinics (pain medicine, surgery, physical medicine) were more likely to write more prescriptions per day but are thought to account for < 20% of all opioid prescriptions per day.   The authors suggest that most of the prescriptions in that category are written by general practitioners, family medicine, internal medicine, and midlevel practitioners.

Overlapping prescriptions ( benzodiazepines and opioids and long-acting and short acting opioids) were common.  Multiple provider episodes or MPEs defined as a resident filling a controlled substance prescription from 5 or more prescribers at 5  or more pharmacies in any 6 month period varied significantly by state, age, and the number of controlled substance schedules added.  The totals ranged from a low of 4.4/100,000 residents in Louisiana to 66.8/100,000 in Ohio.  An overall first impression of this system is that there are limitation but it clearly provides valuable information on prescribing behavior relevant to controlled substances.

The current epidemic of accidental overdose deaths was the driving force for the PBSS system.  It shows that a pharmacosurveillance system is possible, but that there are some limitations.  Data quality as inputted from the pharmacies needs to have quality control measures to assure consistency.  An ideal system would also include a diagnosis or indication.  Physician speciality would also be a useful marker.  I think that the best use of a system like this would be to allow physicians to mine their own prescription data and see how they compare with other physicians in general and within their own speciality.  Specific strategies could be developed for self correction at the earliest possible stages.  I wrote about a pharmacosurveillance system used in Wales in a previous post.  The top 5 drugs causing complications in that system were opioids, antibiotics, warfarin, heparin, and insulin - in that order.  With a sophisticated system it would be possible to pick up significant adverse drug events and monitor those events as well.    

All of the talk about patient safety these days is really about patient safety being practiced in the silos: health care businesses, hospitals, and clinics.  Places where individual health care data is considered proprietary - at least until there is a complication big enough that the state authorities mandate that it be reported for investigation.  These businesses have an inherent conflict of interest in reporting adverse drugs events and severe complications.  Pharmacosurveillance should be out there across the entire health care landscape.  It should not depend on reports about complications made by businesses that are in effect protected by patient privacy.  Complications can be actively sought out and investigated any time a prescription suddenly stops or changes.  This data also needs to be freely available to physicians so that they can look at their prescribing data relative to their peers and make changes where necessary.

It is time to view the process as as a way to learn about how to provide the safest possible environment for patient care,  rather than a way to "catch" somebody when something goes wrong.        

George Dawson, MD, DFAPA


1:  Paulozzi LJ, Mack, KA, Hockenberry JM.  Vital Signs: Variation Among States in Prescribing of Opioid Pain Relievers and Benzodiazepines — United States.   Morbidity and Mortality Weekly Report (MMWR) July 4, 2014 / 63(26): 563-568.

2: Paulozzi LJ, Strickler GK, Kreiner PW, Koris CM.  Controlled Substance Prescribing Patterns — Prescription Behavior Surveillance System, Eight States, 2013. Morbidity and Mortality Weekly Report (MMWR)  October 16, 2015/ 64 (9): 1-14.

From Morbidity and Mortality Weekly Report (MMWR) October 16, 2015/ 64 (9): 1-14.

1 comment:

  1. I'd file this one under great in theory, failing in practice. California's CURES database has been dysfunctional from the outset.

    It would just be a repeat of the EHR debacle.

    I know people in IT and they constantly remind me that health care IT is the Jacksonville Jaguars (yes I know they won yesterday) of IT.