How Databases Can Help Control Antibiotic-Resistant Bacteria
Originally published on Forbes
Nowadays data are being collected and shared on all kinds of things, such as what cat videos you watch, how many cat videos you post on Facebook, and whether you are a cat. But what about data on which patients are carrying antibiotic-resistant bacteria? After all, antimicrobial resistance made the World Health Organization's (WHO's) list of 10 threats to global health in 2019. Would it be helpful to have a database of patients in a region carrying antibiotic-resistant bacteria, so that healthcare facilities can take the proper precautions when such patients get admitted or transfer facilities?
That's the question that our team sought to answer with a study just published in the journal Clinical Infectious Diseases that used a computational simulation model that we built of the Chicago Metropolitan Area and its healthcare facilities and surrounding communities. Different counties and regions have been considering maintaining such a database, otherwise known as a registry, of patients so that public health officials and health facilities could better follow which patients are carrying antibiotic-resistant bacteria such as methicillin-resistant Staphylococcus aureus (otherwise known unaffectionately as MRSA) and carbapenem-resistant Enterobacteriaceae (otherwise known as CRE or really unaffectionately as "the nightmare bacteria"). These bacteria can cause serious infections that are very difficult to treat, because most standard antibiotics no longer work. These bacteria have become too clever, developing mechanisms to protect themselves against the antibiotics, sort of like A.M.A.Z.O. when it was fighting the Flash, Green Arrow, Superman, and Supergirl:
Of course, bacteria aren't exactly like A.M.A.Z.O. There isn't a single giant bacterium, which would be scary, but instead millions and millions of microscopic bacteria. Antibiotic-resistance emerges as some bacteria randomly develop mechanisms to avoid and protect themselves against the antibiotics. These bacteria are then hardier and are thus better able to reproduce, resulting in more bacteria in a population that are resistant to antibiotics.
Therefore, infection control specialists in hospitals often try to keep patients who have these bacteria separate from others so that they won't spread the bacteria to other patients. In some cases, they may also bathe such patients in antiseptics to try to get rid of the bacteria.
However, to keep such bacteria from spreading, infection control specialists have to know which patients carry such bacteria. Such bacteria tend to be very quiet until they cause an infection. They don't tend to throw raucous parties, play loud music, and yell "YOLO," while they are simply "colonizing" you. You may not know that you have such bacteria on your skin, in your armpits, in your groin area, or in your intestines. But when they start causing skin infections, urinary tract infections, pneumonia, bloodstream infections, and other sorts of problems, they can be a whole load of trouble. Only a limited number of antibiotics now remain to treat such infections. The concern is that soon even these antibiotics will not be effective. These bacteria are fast learners and our society ain't producing new antibiotics fast enough.
Moreover, each healthcare facility isn't like Gilligan's Island or that island on the television show Lost. Instead, as our previous network analysis published in the American Journal of Public Health showed, healthcare facilities in a region can be highly interconnected via patient sharing. A patient who leaves one facility may enter another one, either directly via a transfer, or after staying at home for a while. There are lots of reasons why this may happen. For example, some patients frequently have to change health insurances because they change jobs or lose coverage and therefore need to seek care at different facilities. Alternatively, their doctors may change facilities. There are also patients who doctor shop or need certain services that aren't available at a present facility. Thus, it is not surprising that patients who pick up antibiotic resistant bacteria in one facility can then spread it to other facilities. Our studies published in Medical Care and Infection Control and Hospital Epidemiology (often referred to by its acronym ICHE, which has the delightful pronunciation "itchy") showed how quickly an antibiotic-resistant bacteria outbreak in one facility can spread to nearly all of the other facilities in a region. Therefore, it could be helpful for the different facilities to communicate with each other and public health officials. In fact, our study published in Health Affairs that used a simulation model of Orange County, California, showed that when different health care facilities coordinate infection control measures, they can better prevent and control the spread of antibiotic-resistant bacteria.
All of this suggests that maintaining a database to know who is carrying antibiotic-resistant bacteria in what facility may be helpful. But how do you test this possibility? You could just try it, but that could cost a fair amount of time, effort, and money if it didn't work out well. Plus, many questions remain about the design of such a registry, such as how many facilities need to participate and agree to share information for the registry to be effective. Certainly, just one or two facilities among dozens or hundreds of facilities would not be enough. That would be like only one or two people reviewing restaurants on Yelp.
Enter computational simulation modeling. In other fields like meteorology, air traffic control, aerospace engineering, and finance, they often model and simulate what may happen before trying it in real life. This can save money, time, and effort and prevent potential big mistakes. It also offers the opportunity to refine the strategy. Of course, computational simulation modeling is not perfect and doesn't preclude the need for other types of studies, but it can offer insight that otherwise can't be gained.
For the study, our team developed a computational simulation model of 90 acute care hospitals, 9 long-term acute care hospitals, 351 skilled nursing facilities, and 12 ventilator-capable skilled nursing facilities in the Chicago metropolitan area and the many, many, many patients moving between the surrounding communities and these facilities. This was an agent-based model and still is, not because it was created by secret agents or included only patients who were talent agents, sports agents, or Agents of SHIELD. Rather, this model represented each patient with a computational agent, a virtual person who could make his or her own decisions (i.e., exhibit autonomous decision making) and exhibit behaviors that change with time and experience (i.e., complex adaptive behavior). This simulation model then could serve as a "Virtual Laboratory" to test what could happen if you tried new policies, interventions, and strategies, such as implementing a patient registry that tracked antibiotic-resistant bacteria.
Our team and authors for the study included members of our Public Health Computational and Operations Research (PHICOR) modeling team at the Johns Hopkins Bloomberg School of Public Health (Sarah M. Bartsch, MPH, Leslie E. Mueller, MPH, and me), the Pittsburgh Supercomputing Center/Carnegie Mellon University (Joel Welling, PhD, Jim Leonard, Jay V DePasse, and Shawn T. Brown, PhD), and Rush University (Mary K. Hayden, MD Sarah K. Kemble, MD, Robert A. Weinstein, MD, Kruti Doshi, MBA, William E. Trick, MD, and Michael Y. Lin, MD).
For this study specifically, we focused on CRE, not just because it is nicknamed the "nightmare bacteria" but also because it has been a growing problem in the Chicago-area and members of the Rush University team are world's experts on this bacteria. Here's a CBS News segment on CRE:
Using the model, we ran different scenarios simulating the use of different types of registries and varying the number of healthcare facilities who were willing to share information and check the registry when patients were admitted to their facility. For example, a registry may send an electronic alert to the responsible doctor if a new patient admitted to a health care facility is carrying CRE. We found that even when only 25 percent of the largest health care facilities used such a registry, the number of new CRE carriers in the Chicago Metropolitan Area went down by 9.1 percent over a three year period. Bump up the number of Illinois healthcare facilities participating to half and this number increased to 10.7 percent while the overall prevalence of CRE prevalence dropped by 5.6 percent. Having all 402 Illinois health care facilities participating increased these decreases to 11.7 percent and 7.6 percent, respectively. Again, this all occurred because infection control specialists could then know quickly when a new patient was carrying CRE and then take proper precautions like separating the patient from others and having healthcare personnel wear protective gloves and gowns when handling the patient.
“Identifying CRE patients with a registry can save considerable resources and time, especially if known carriers do not need to be re-identified,” said Bartsch. “Thus, a registry for extensively drug-resistant organisms could be an effective tool in combating the spread of antibiotic-resistant bacteria between health care facilities in a region.”
Sounds like a positive story for building and maintaining such a registry? Perhaps. Of course, our model and study were not perfect and had limitations. We didn't have data (such as the prevalence and incidence of CRE) on every single facility so had to make some assumptions about facilities where data was lacking. We used Medicare and Medicaid data to determine where patients may move and not other types of insurance data (e.g., commercial insurance). This assumes that patients with other insurance and no insurance may move in the same patterns. A computational model certainly cannot represent every possible factor and thing that can occur in real life.
In real life, such databases are not perfect either. People may enter the wrong information or forget to check the database when a patient is admitted. They may not enact precautions quickly or carefully enough either. Furthermore, it is important to remember that a model won't necessarily give you exact predictions but it gives you a sense of what may happen and the factors that may affect the results.
Moreover, healthcare facilities may be reluctant to participate in such a registry. They may not want to share which patients are carrying antibiotic-resistant bacteria because it may not sound like a selling point for the facility. Additionally, any information on individual patients need to remain protected for the patient's sake and used only to help patients. You don't want a situation where you start getting targeted advertisements for special antibiotic-resistant bacteria hocus-pocus supplements and hats (which wouldn't be helpful).
Nevertheless, with the field of data science burgeoning, this study does show how data and data approaches could help combat antibiotic-resistant bacteria. We will still need new antibiotics and do other things such as control the widespread overuse of antibiotics. But here's an area where better data and data approaches can do more than determine whom to send cat product ads.