It is September 3, 1854 and 70 people from a single London neighborhood have died from diarrhea in the last 24 hours. If they died today, doctors would identify the cause of death as “vibrio cholerae”, but the residents of Soho won’t know the bacterial cause of cholera for another 30 years. The prevailing medical wisdom in 1854 is that cholera is caused by miasma, or “bad air”, and common sense wisdom is telling Londoners to leave the neighborhood as fast as possible. One man (well actually two) thought and acted differently. From Steven Johnson’s fantastic book on the Broad Street Cholera Outbreak, The Ghost Map:
“But one Soho regular had been following the case closely from his residence at Sackville Street on the southwestern edge of the neighborhood. Sometime near dusk he set out from his home, marching through the empty streets, directly into the heart of the outbreak. When he reached 40 Broad, he stopped and examined the pump for a few minutes in the fading light. He drew a bottle of water from the well, stared at it for a few seconds, then turned and made his way back to Sackville Street.”
That man was Dr. John Snow and that pump is the Broad Street pump. Dr. Snow went against the medical establishment and the flow of people out of Soho and into the annals of medical history, discovering the water-borne transmission of cholera by using data as we aspire to today, while teaching us a lesson largely buried by our molecular understanding of medicine --the cause of death was a water pump.
Fast forward now to 1978 and the pseudonymous Dr. Samuel Shem publishes The House of God, a scandalous satirical novel about medical interns trying to survive their Boston training. Described by the New York Times as “raunchy, troubling and hilarious”--and attacked by the medical establishment of the day--its portrayal of physicians is less than flattering. And yet 40 years later, The House of God, remains part of the discussion because it is in our discussions--the novel introduced an entire lexicon of gallows humor vocabulary that one can still hear on inpatient wards today. One of those terms--Mount Saint Elsewhere--is an antonym for the Broad Street pump.
Mount Saint Elsewhere is a generic, fictional, sender of transfer patients to an academic hospital. It works like this: Question: “Where did this difficult patient come from?” Answer: “Mount Saint Elsewhere.” The term captured an unmentioned truth about medicine and became the title of a popular t.v. show in the process: If you are on the receiving end of transfer patients, it is tempting to think of every hospital sending you patients as a vague and congealed force.
I thought of the Broad Street Pump and Mount Saint Elsewhere looking at our own data from two neonatal speciality centers in the eastern Congo. The mortality rates of babies transferred into the speciality centers after birth is much, much higher than babies with the same diagnoses, born at the speciality centers themselves. After an informal survey of the literature, we are not alone. I found examples from Uganda, which directed me to other examples in Nigeria, Cameroon and outside of Africa in Bangladesh. In other papers, outborn babies were conspicuous by their exclusion.
There are many hypotheses that make intuitive sense to explain this finding in low resource settings. Here are two:
But these hypotheses assume an optimal flow of babies. Let’s define optimal flow. In any system designed to care for babies, babies move by design. This is true for two reasons. First, not every clinic is capable of caring for all sick babies--to maximize efficiency there is a necessary concentration of expertise and equipment at speciality centers. Second, we don’t always know which babies will be sick in advance or, if we do suspect a sick newborn, the mother cannot always be transferred before delivery. With this as background, an optimal flow of babies is one where:
Not surprisingly, I was unable to find authors who concluded that higher mortality in low resource settings among babies transferred from other facilities is what one would expect and thus not a cause for concern. The Uganda study and others come to the opposite conclusion. Speaking to our work in eastern Congo, there are four reasons why I strongly suspect we are not in an optimal flow system: (1) the magnitude of the mortality of outborn babies (2) their condition on arrival (3) the fact that many are discharged home and then “bounce back” (another House of God term!) to tertiary facilities and (4) the presence of treatable, and even preventable, admission diagnoses.
But here’s where I think we must do better. We can’t be satisfied with “outborn” as another Mt. St. Elsewhere. So what is preventing us? There is a paradox in neonatal care--in both strong and weak health systems the movement of babies looks identical, sick babies move from low-level facilities to speciality centers. In strong health systems this is a feature--maximizing care through the concentration of resources. In poor health systems this movement can mask a bug--low quality care and lack of training leading to illness that requires care escalation. To trace and reduce neonatal mortality, we must differentiate instances where the movement of a baby worked as designed, from instances where illness may have been prevented or mitigated at the birth site. What we are missing is the data that puts the movement of babies in context. With a generic “outborn” label, we are left trying to understand the complex movement of extremely high-risk babies across a geography by looking only at their arrival destinations.
With our NoviGuide software we are starting to collect this data, and more, as a result of trying to solve a completely different problem. The NoviGuide is a clinical decision support software that guides bedside clinicians through the care of a newborn. While midwives liked the NoviGuide, they told us that they didn’t like having to tell NoviGuide what they had available to treat babies every time they assessed a newborn. For example, if midwives couldn’t measure a glucose level because they didn’t have a glucometer they thought NoviGuide should “know” that by the 20th baby and stop asking. We fixed this by having sites enter in their available resources when NoviGuide is first implemented and then configuring guidance to a site’s specific resource profile (we’ll explain how we did this in a future blog). Hospitals can always update the list if new equipment arrives or old equipment breaks.
By knowing what hospitals have, NoviGuide can tailor guidance and keep nurses engaged. But its real power is in generating system level data. Knowing what resources hospitals have at baseline, and then comparing expected resource utilization to NoviGuide usage data, helps administrators know when something is missing, broken or not being used. It also helps health administrators recognize when the baseline itself needs to be raised by comparing what sites have to what international standards tell us they should have. If we want to know when the movement of babies is working “as designed” we at least now have the blueprints.
With NoviGuide implemented across an ecosystem of hospitals, we can know what hospitals have to treat babies and what they tell the NoviGuide about how they use those resources. If a baby gets transferred, it’s also feasible to know where they went and what happened to them once they got there because transfers are generally sent to higher capacity sites with dedicated neonatal units and better record keeping about neonatal outcomes. It is a start, but not enough, for two reasons. First, to reduce mortality we need to not only know about the babies that were transferred, but also those who maybe should have been transferred and were not. Second, unlike Dr. John Snow who ended the cholera epidemic by convincing the health board to shut off the pump, we are going to need to fix the pumps--hundreds, even thousands, of pumps with distinct problems, and on a tight budget. To do that we are going to need data tools that can find the hidden correlations, adapt easily to new data, be transplanted to any ecosystem of birth sites and continuously improve with each transfer baby data point as dynamic interventions are underway.
Babies do not die simply from having diseases. If they did, then mortality would parallel pathology and be uniform around the globe. A baby dies from having a disease in a specific place, not a generic “elsewhere”, but a distinct location with dynamic human resources and technologies to identify and care for sick babies, and a fixed proximity to more advanced care. There is a lot we need to learn about sick babies in low-resource settings. We can start by using data to travel the same paths babies do.
