By Dr. Cole Stanley
Turning diagnostic coding from a necessary evil into a vital practice tool
I can think of few things that spark less joy in medicine than the task of choosing the proper diagnostic code(s). Too often I find myself on page 3+ of the search results, wading through the overly detailed and outdated nomenclature, still coming up empty. Imagine for a moment this entirely plausible scenario: One of your Gen Z clients, a self-described “influencer”, has a large goose egg on their forehead. Inconsolable, they shed tears down on their cracked black mirror. Meanwhile, you’re busy staring at your EMR- what’s the code for “injuries related to walking while TikToking”? How about this one – “W220.2XD: Walked into lamppost, subsequent encounter”? Not quite right, as happily this isn’t a recurrence. Maybe the underlying cause – “12.Z99.89: Dependence on enabling machines and devices, not elsewhere classified”? At that very moment, you become aware of the coding rabbit hole you’ve again fallen into, and with a sigh – “780 – General Symptoms” it is. (P.S. these are all actual ICD-9/10 codes)
Many of us throw our hands up in the air and settle for the generic – enter 50B anxiety/depression, 780 general symptoms and the like. These generic codes consistently rank near the top across many settings and jurisdictions. A simulation of coding in two systems in the US showed that only about half of included codes were “appropriate”, and about a quarter were omitted (source link). I suspect many rightly feel that the time spent getting the billing coding exactly right could be much better spent on other work (i.e. things that we actually perceive as contributing to the quality of care we provide). The resulting situation: wasted time, feelings of not performing to standards, decreased sense of self-efficacy, and a general aversion to the whole task of coding. We start dreaming of being able to simply write or dictate our notes, and then letting a third-party or technology handle the rest – “just let us practice, already!” But I’m writing this today as I’m worried about what we may be giving up. Instead of outsourcing, we need to embrace coded health conditions as a vital practice tool. I’ll share the reasons why we need to do this, what we risk if we don’t, and how you can get started.
Let’s start with the paradigm shift we need – the primary purpose of coding health conditions should NOT be for billing. Instead, we should see it as a necessary tool for practice management. Physicians and their teams are taking this up across the province with the help of PSP’s panel management program, but there is still work to be done. HDC data shows that only 55% of clients with an encounter in the past year had a coded health condition entered in their EMR problem list.
A team in Amsterdam found something similar, with widespread variability in problem list use (source link). One of their key recommendations was that peer-to-peer teaching on the benefits of using the problem list is needed, and herein lies my attempt at this.
In my last article, I tried to make a case for why we need to embed QI efforts in our frontline work, and the risks of a widening “Know-Do gap” if we don’t. If we leave accurate coding of health conditions to post-hoc third-party efforts, I fear we’ll similarly be widening this gap. To understand our practice panels, we’ll need to rely on complex algorithms, natural language processing (NLP), and potentially Artificial Intelligence (AI) that “guess” at conditions our patients may have. These strategies can work well if we want to get the most accurate picture of what is going on in our practice right now (before data clean-up), but aren’t as well suited for our own QI efforts.
I think back to the example of our BOOST Collaborative, where we aimed to improved retention on opioid agonist therapy (OAT) for our clients with opioid use disorder (OUD). If I simply wanted to get a best guess at the prevalence of OUD in my panel, I might rely on an algorithm that uses keywords, text strings, prescribed meds, or even natural language processing – no need for coding right? But what about the inevitable false positives and negatives? Physicians and their teams are quick to find faults in data, lose trust, and discount its utility for QI. They’d point out that some clients who they know have OUD are missing, and others on the list don’t actually have OUD (e.g. “they’re only on that methadone for cancer-related pain”). The QI team wouldn’t want to use that list for patient-specific tracking or interventions. Instead, they’d prefer a simple inclusion definition less prone to false positives and negatives. In QI, we call this determining the population of focus (POF).
Part of the fun of doing QI is that we are afforded the chance to look to analogous problems and their solutions in places outside of healthcare. When I think of the messy state of coding in our EMRs, I guess it’s no wonder that Marie Kondo popped into mind. Her rules for tidying up (KonMari MethodTM) hold some useful advice we can integrate with our QI lens to create this list:
Principle | An example from my practice | |
1 | The Why – Get clear with your team on the benefits of coding health conditions in problem lists | Our Hope to Health clinic team embeds QI in our regular work, and so our top reason for coding health conditions is that we need to establish reliable populations of focus (POFs) |
2 | Start small – Choose a single condition that is important to you and your team | We decided to first focus on opioid use disorder. (Prevalence is very high in our population and we devote significant QI efforts to this) |
3 | Choose the best code – look at the ICD-9 coding resource, or Notes section on relevant measures in HDC Discover | We choose the ICD-9 code 304.0 for opioid use disorder (OUD), similar to what was used for the BOOST Collaborative). |
4 | Tidy up – use EMR queries to find clients that likely have the condition but are not properly coded (and see #6 below) | We created EMR clean-up tools that showed clients with suspected OUD but no coded 304.0 (based on prescriptions for possible OAT) and added the code where appropriate |
5 | Make the right choice the easy choice – use EMR functionality (e.g. favourites) to more easily add appropriate codes to problem lists and simplify list of options | We added 304.0 as a favourite in our EMR so it can be easily added to the Problem List. |
6 | Get help – HDC team members or PSP coaches are available, or look to local champions | We rely on the team’s continued QI efforts and embedded clinical QI leadership to maintain the above |
For BOOST, we decided that we’d use 304.0, and used queries to identify clients and add this standard code (here’s where NLP and AI could be useful in this initial clean-up). We also built in a quick-add button for the code on our OUD form in the EMR. Where this was implemented, teams are able to understand and trust a list of their clients with OUD, as we’ve fixed the “garbage in, garbage out” issue. I’m advocating for us to add a prevalence measure for OUD to HDC Discover, while knowing that this first step of determining the POF will be needed for groups to find it useful.
Determining the POF does take time and effort, but it feels worthwhile when we see what the result can be used for. In BOOST, teams ended up with reliable data for their patients’ retention on OAT, and could use the list from their EMR for follow-up efforts. The benefits don’t stop there though – see the table below. Many of these touch on the economies of scale we can access when we incorporate panel management (vs. solely “one patient at a time” care).
Benefits of diagnostic codding as a vital practice tool:
- Frontline trust in the data: Improves identification of population of focus (PoF) for QI efforts
- Timeliness of data: Real-time, with little post-hoc processing required
- Accuracy of QI measures: Accurate after initial data clean-up (determining PoF) if good ongoing documentation practices, false positives and negatives less of a worry
- EMR functionality: Can use Clinical Decision Support (CDS) features of the EMR to do proactive recalls, group medical visits, reminders, etc.
- Quality monitoring: Gives the team a reliable patient list to review for focused efforts (i.e. periodic team review of PoF patients)
- Documentation and billing: Coded conditions can be quickly added to documentation or bills with EMR shortcuts, saving time
- Specialist referrals: Coded problem lists can automatically be added to referral letters, saving time.
- Measuring complexity: Coded problems lists can be used to estimate the complexity of your patient panel
- Measuring prevalence: Coded problem lists can help us more accurately measures prevalence, which can be used for resource allocation (i.e. community X needs “insert health professional here”)
Got an idea that needs a sounding board? Or help to decide which ICD-9 code to choose? Contact me. I’d be happy to help or pull our HDC team into the conversation if needed. Happy coding!