Inference in the Pharmaceutical Supply Chain

In my recent essay, “RFID is DEAD…at Unit-level in Pharma”, I used relative cost estimates to theorize that Radio Frequency IDentification (RFID) will not be the predominant carrier of serial numbers applied to drug packages for compliance with U.S. pedigree laws.  My theory is that there will actually be a mix of RFID and 2D barcodes, and that barcodes will be the predominant carrier technology.  As you might expect, that was a little controversial with the RFID industry (See RFID Journal’s blog post in response to my essay).  Hey, it’s just a theory.  I too, wish everything would be RFID, and for all of the reasons cited by RFID Journal.  It’s just that I don’t believe it’s going to turn out that way and that’s based on the logic I laid out in my essay.  (Also see FiercePharma Manufacturing’s more neutral post in response to my essay.)


In a mixed-but-predominantly-barcode-serialized U.S. pharmaceutical supply chain, companies will not be able to tell which unit serial numbers are inside of each sealed case at receiving or at shipping.  Because barcodes are a “line-of-sight” technology, the only way to tell with absolute certainty what the sealed-up unit serial numbers are will be to cut the tape seal, open the case, expose the barcodes, scan each unit and tape the case closed again.

It is a ridiculous notion to believe that this should happen at each receiving and shipping point because it would measurably elevate the risk of error introduction, damage and theft.  It would even cause greater suspicion of tampering by downstream trading partners who receive cases that have obviously been cut open and re-sealed after the manufacturer originally sealed them.  But most significantly, it cannot happen because it is super-inefficient (read:  unacceptably costly).

Notice here that if the units had been serialized with RFID tags it would not be necessary to open the case to read the unit serial number so each recipient of the case could read the contents without opening it.  Unlike barcodes, RFID does not require “line-of-sight”.  But because barcodes do, a recipient can never be sure what is hidden from sight inside a case without opening it up.

This is a well-known dilemma in the industry groups that I participate in.  The California pedigree law requires recipients of drugs to certify, under some stiff penalty, that they received specific drug units based on the serial numbers.  That’s how a pedigree works, but to be able to really certify that you have received a given set of unit serial numbers that only use barcodes, you would have to open, scan the units and reseal every case.  But if this can’t happen without a huge increase in costs, the only other way to do it is to simply trust what your upstream trading partner told you was in each case.  After all, the same pedigree law applies to them.  They are required, under the same stiff penalty, to give you a pedigree with exactly the serial numbers that they shipped to you so you should be able to trust them, right?

As long as the manufacturer puts a serial number on each case, whether RFID or barcode, and they read each of the unit serial numbers, whether RFID or barcode, as they are loading them into the cases, and if they accurately associate the case serial number with the unit serial numbers contained inside,  they would build a containment hierarchy for each case.  This is perfect for an Electronic Product Code Information Services (EPCIS) aggregation event, or a Drug Pedigree Messaging Standard (DPMS) envelope.  Both of these XML documents are designed to hold the kind of serialized containment hierarchy that I just described.

Armed with the containment hierarchy documented in this way, one could infer the unit serial numbers contained inside of a case by using the case serial number as a lookup key.  This is known as “inference” in the pharmaceutical supply chain.

Inference can be performed at any level that the containment hierarchy allows.  If the containment hierarchy includes unit-to-bundle, bundle-to-case, case-to-pallet, pallet-to-shipping-container aggregations, then given a shipping container serial number and using inference, one could determine (infer) all of the unit serial numbers that are present inside of it without opening and without scanning(barcode or RFID).


The topic of inference in the drug supply chain has been getting a lot of attention in the Traceability Adoption work group of GS1 Healthcare U.S..  That group just published their paper, “The Practice of Inference in the U.S. Pharmaceutical Supply Chain Pedigree Regulatory Environment” which provides a good definition and background on the concept.  My purpose is not to reproduce the contents of the GS1 paper but, just to make sure you know what we’re talking about here I’ll give you my own definition of inference.

Inference is an approach to identifying the contents of a shipping container without actually opening it to confirm exactly what it contains.  The types of “shipping containers” most often considered for the application of inference are cases packed by manufacturers and totes packed by wholesalers.  Being “an approach…”, the use of inference by any given company within any given business process (normally shipping and receiving), is optional and would be a conscious decision.

When units are serialized with barcodes, if you choose not to use inference then you simply must open every container and confirm the contents by viewing and/or reading them.  Use of RFID serialization on items inside the container could be used to avoid the use of inference and still not open the container, but that assumes that you will always read 100% of the tags.  As soon as you get anything less than 100% you will have to decide whether you want to open the container to confirm the presence of those units you didn’t read, or use inference to infer that they are present.

There is an important distinction between the use of inference with serialized and non-serialized items.  Today, very few pharmaceuticals have serial numbers applied (either barcode or RFID) at the item level.  When trading partners ship and receive a full casepack of drugs they normally never open it just to confirm that it contains exactly what the case label and the ASN says.  They infer that the case label is accurate, including NDC and quantity.  This type of inference is standard operating procedure across the entire supply chain—and has been from the beginning of time.  Use of inference specifically for serialized units is the focus of the current brainstorming activity.  This type of inference allows one to infer a given container contains the NDC and quantity, and the exact serial numbers that are specified in an ASN or pedigree.


You can’t really understand inference unless you understand exactly why it might be needed.  The need is the result of the combination of three important things that I have previously written about:

  1. The California Pedigree Law (effective in 2015-2017) will require shippers and receivers to certify, under penalty of fines and potentially prison, that the exact item serial numbers inside of the shipping containers match the pedigree;
  2. The modern U.S. pharmaceutical supply chain moves massive quantities of drugs at a high level of efficiency on a daily, overnight delivery cycle;
  3. Due to the continuing relative high cost of item-level RFID, most Pharma manufacturers appear to be planning to use 2D barcodes to serialize their items.

The California Pedigree Law has a very noble purpose—to protect the public from crimes perpetrated against the pharmaceutical supply chain and ultimately against patients.  I don’t question that purpose.  I know the purpose of the pedigree law is not to destroy the efficiency of the current supply chain, but without the use of inference, it will.

When we reach the point where many pallets-worth of serialized/pedigreed products are moving through the supply chain (unlike today where it’s just a few SKU’s and the shipment quantities are small), trading partners will be forced to maximize the use of inference whenever they can to avoid having to open every case.


But there are problems with the use of inference.  First, no pedigree law automatically allows its use.  Supply chain members who plan to use inference may find that law enforcement officers will not accept the excuse that they were using inference whenever the actual serial numbers in a shipping container do not match the pedigree.

Though rare, mistakes in casepacking, shipping and receiving occasionally happen today.  Even at the low occurrence frequency of these errors, the vast majority, if not all, of them are unrelated to the crimes that are targeted by the California Pedigree Law.  They are just unintended mistakes.  Today, the primary driver for trading partners to correct these errors is financial.  Once the law goes into effect, mistakes like these will open each occurrence to wider scrutiny and potentially result in a fine or worse.

Second, the decision by a member of the supply chain to use inference must be based on their trust that the company who packed the container has very reliable case/tote packing and data collection processes.  Think about what will happen under the California pedigree law when the contents of a single case are not captured properly by the manufacturer (or their contract packager).  That is, the containment hierarchy for the case that is supplied to the wholesaler is incorrect.  Most wholesalers open the majority of the cases they receive at some point during their processes, but a few cases are shipped to customers without being opened.  Imagine that this incorrectly documented case is one of those that are passed on to their customer.  When using inference, the wholesaler will simply pass on the same containment hierarchy they received from the manufacturer.

Now assume the wholesaler’s customer opens the case and finally finds that the serial number on one or more units are not the same as the electronic pedigree record indicates.  Those units have no value regardless of the price of the drug because the owner does not possess a valid pedigree and they cannot be sold without one.  There are only three ways to address this situation:

  1. Destroy the units and take a total loss on the undocumented units;
  2. Contact the wholesaler and negotiate with them to get a valid pedigree from the manufacturer and pass it to them.  This is unlikely to ever work because neither the wholesaler nor the manufacturer may have anything that confirms that they made a mistake.  Even if the greater mess gets revealed to them when someone else makes a claim that they received the matching undocumented serial number (which is far from guaranteed when only a single state has a serialized pedigree law), the manufacturer may not feel confident enough to update their original pedigree document and pass it to the wholesaler.  If they do, then the wholesaler also has to feel comfortable updating that new pedigree with shipping information that occurred in the past and finally pass it to the current owner;
  3. Contact the wholesaler and negotiate a return of the undocumented units for credit.  The wholesaler may not be receptive to taking the undocumented units back because they have nothing to indicate that an inference error really did occur.  They may fear that they will get stuck with the unit and end up having to take a loss on it before it gets resolved.

All of these alternatives represent a lot of special handling and cost.

In addition to these special handling costs, inference errors will cause shortages of product in the supply chain.  That’s because the supply chain is becoming increasingly optimized through lean approaches to inventory management.  When a case of drugs arrives at a location in the supply chain, it is needed to fulfill customer orders, including patient prescriptions.  If even a handful of drugs are ordered, shipped and arrive but cannot be sold because of an inference error, chances are order fulfillment of some kind could be delayed.


Surprisingly, the Ridley Thomas bill in California (SB 1307) that became law in the fall of 2008 included more detail about inference than most people realize.  See Section 4163.3 of the California Business And Professions Code, which reads, in part:

“…the board shall, by regulation, define the circumstances under which participants in the distribution chain may infer the contents of a case, pallet, or other aggregate of individual units, packages, or containers of dangerous drugs, from a unique identifier associated with the case, pallet, or other aggregate, without opening each case, pallet, or other aggregate or otherwise individually validating each unit.”

It goes on to say:

“Manufacturers, wholesalers, and pharmacies opting to employ the use of inference as authorized by the board to comply with the pedigree requirements shall document their processes and procedures in their standard operating procedures (SOPs) and shall make those SOPs available for board review.  SOPs regarding inference shall include a process for statistically sampling the accuracy of information sent with inbound product.”


“Liability associated with accuracy of product information and pedigree using inference shall be specified in the board’s regulations.”

As I understand it, the Board of Pharmacy has deferred their mandated work on inference until we get closer to the effective date of the regulation.


In my last post, “The Modern Pharmaceutical Wholesaler and the Approaching Transformation” I said that I had observed a significant shift in attitudes creeping in about how a serialization-based supply chain pedigree system might operate.  I said it I thought it was a direct result of the growing realization that RFID will not be the predominant carrier technology at the unit level.  A few important manufactures are realizing that their choice of adopting barcodes instead of RFID will force their downstream trading partners in the supply chain to rely too much on the accuracy of their containment hierarchy information for use in inference.  This essay is already way too long, so in my next post I will analyze this concern and explore where it leads.

3 thoughts on “Inference in the Pharmaceutical Supply Chain”

  1. Dirk – Interesting article and blog.

    If your assumption is correct and the de facto serialization is 2D barcode, then the choice is either inference or reconsidering – and going back to non-line-of-sight (RFID).

    On RFID for a moment, I am curious as to what readability rates are at this point on units. Even at 98-99%, if you think about the number of units per case, plus the case itself – running through wholesalers at high speeds, there would be very frequent stops in process at those seemingly-high readability rates. Different pilots have yielded different results – I’d be interested if anyone has had better than 99% on RFID.

    If, as you suggest, we take 2D barcode / inference as a given, then, I have to ask if the additional cost of unit level serialization provides any dramatic safety benefit that case-level with a parent-child relationship between manufacturers cases and repacked totes, etc. wouldn’t. In short – inference or consideration of RFID are issues because of unit-level serialization – is the additional cost and confusion worth it in terms of the safety results we can anticipate as a result?

    It would be great if California Board of Pharmacy would take up inference as well as other rulemaking now – as the supply chain plans compliance – rather than “closer to the implementation date”, when it might change some of those plans late in the game. The more stable and sure the regulations are now, the less risk in the investment – especially for smaller manufacturers and others in the supply chain.

    1. Brian,
      Your comments are excellent points. First, you may be right that the only option to 2D barcodes and inference is reconsidering RFID, but stay tuned for my next installment. I can’t share with you any read rates, but I can suggest to you that the “RFID read rate” of a case of 2D barcoded units is 0% and inference would be required on 100% of the units. On the other hand, an RFID read rate of a case of RFID-tagged units that is, say, 97% (a pretty bad read rate for RFID) means that inference would be required on only 3% of the units. That’s a huge difference. Now that’s an inference rate that regulators might be able to accept with only occassional sampling and verifying. To me, that’s still workable.

      I think the additional safety benefits of unit-level serialization, over case-only-serialization are significant in the U.S.. The majority of cased in this supply chain only travel from the manufacturer to the first wholesaler. Beyond that, the first wholesaler places the units into serialized totes (not all wholesalers serialized all totes) and ship them to pharmacies and, in some instances, to other wholesalers. What I think you are describing is a situation that exists today in Florida, where wholesalers must keep track of the supply chain history of each case, and then carefully keep a connection between the source case and the destination tote without the benefit of unit serialization. One approach is for the wholesaler to add their own serial number to each case as it is received so that they can trace its history easier, which would be similar to what could be done if the cases were serialized by the manufacturer instead. I can tell you that this approach has so many holes in it that it should be considered to be totally unworkable in the U.S. supply chain. It is way too easy to lose the trace of a given unit, and without unit serialization, it would be way too easy for criminals to “cleanse” the history of a given unit undetectably. See my post on the Florida Pedigree Law:

      I have come to the same conclusion that you have about the need for the Calfornia Board of Pharmacy to take up the issue of inference sooner rather than later because it could have a huge impact on the carrier technology selections of pharma manufacturers. These are choices that are being cast now and will be very difficult for mid- to larger companies to change even in 2013, let alone 2015. Actually, I think the smaller manufacturers are lucky in that they will be able to start their deployments later and will be able to take advantage of knowledge that isn’t available yet.

      Thanks for the excellent dialog. These are exactly the questions and issues that we need to get out in the open and debate/discuss now, to help the industry make better decisions about adoption. The RFID/barcode/inference debate is one of the most important.


  2. Dirk – Good dialog. A few additional points on the RFID front.

    1) When you mention RFID, do you mean UHF Gen2 vs HF? In the earlier RFID testing cycles, one of the issues was which protocol. This would need to be firmed up as read cycles are proven so the reader and tag environment would be consistent across the supply chain.

    2) To get better reads could mean more RF power, as one tuning factor of many. This can also mean spurious reads of other tags, which would need to managed in the RFID environment. Shielded tunnels and filters may be necessary, including RF site surveys for interference, etc.
    This should not be overlooked.

    3) There was also a limitation of mobile devices that cleanly worked with 1D, 2D barcode and Gen2 RFID reads. This was another factor that limited the easy adoption of a seamless RFID solution working with secondary barcode backup capability.

    Though it may not show from the above comments, I am a strong RFID advocate and would like to see the x-ray of cases versus inference of barcoded labels in cases. I certainly have seen a successful full x-ray of cases as well as full x-ray of pallets. But consistent read rates across product composition (glass, liquid, syringe, powder, pill, metal lids, etc.) and a proven set of technologies to show RFID working in a seamless manner as barcodes are used today are other factors that will help drive RFID adoption further in Pharma.

    There are no doubt efficiencies to be gained by RFID. And inference is not the best policy but agree will be required for 2D barcoded units in a case. The additional cost of tags when multiplied by product volumes as well as the cost to encode RFID tags onto units including aggregation are significant additional factors that drive back to 2D barcodes at this point in time, from my opinion. Time for the board to step up the definition of inference rules if 2D barcodes will be the reality. Though an area, they may be hesitant to define…

    I do believe RFID is the future path, the question is when and also what areas it can be used today to provide immediate value.

    Hope I an not crossing into the next post… 🙂

Comments are closed.