Close Proximity Interaction and S. aureus Spread in Long-Term Care

There is little doubt that S. aureus is transmitted between patients via contaminated hands, white coats or fomites carried by healthcare workers. Fortunately, that doesn't curb the enthusiasm of scientists seeking to understand the mechanisms of transmission in clinical settings. A case in point is a recent study in PLOS Computational Biology by Thomas Obadia and colleagues that tracked healthcare worker (HCW) and patient close proximity interactions (CPI) via small wireless sensors and correlated the interactions to incident S. aureus colonization.

The study was conducted in a 200-bed LTCF in France and utilized data collected from 329 patients and 261 HCWs over a 4-month period. Using weekly nasal swabs from all patients and HCW, they spa-typed and determined the resistance profiles to antibiotics of each detected S. aureus. Isolates were considered identical if they shared the same spa type and resistance profile. Incident cases were only considered in patients since HCWs could have been transiently colonized and missed by the weekly swab interval. Each incident case was evaluated to confirm that at least one time-consistent CPI in the prior three weeks with the same strain was possible. The 3-week interval was chosen since all but 4 cases could be linked to another patient with the same strain during that time period.

Without going into all of the statistical modeling methods, major findings include:

(1) When limiting the analysis to the 201 patients not already colonized on admission, 73 acquired S. aureus. The one-month acquisition rate was 33%.

(2) Time to acquisition among new admissions did not change based on the number of colonized neighbors in the preceding week using either raw number of CPIs nor cumulative duration of CPIs.

(3) There were 237 incident cases throughout the 4 months in 111 patients. Only 173 had candidate transmitters. The analysis was limited to 153 because sensors failed to record a CPI in the prior week in 20 cases.

(4) A CPI path existed for 149 of 153 episodes. As seen in the example figure above, P1 (patient) and H1 (HCW) were two-hops away from the incident case. P2 was three-hops away. In the same figure, distance between a transmitter and an incident case (black) was shorter than random simulations predicted (white). Additionally, a direct contact between candidate transmitter and incident case (i.e one hop away) occurred in 48% of cases vs. 30% expected by random chance. These findings supported that CPI predicts S. aureus incident colonization.

(5) HCWs spent about 20% of their shifts in direct contact with patients (110 minutes over 8 hours) and had CPIs with 15 unique individuals during their day (9 were patients, 6 other HCW) with 3.7 hours spent in contact with others. 36.3% of HCW were colonized with S. aureus.

(6) Interestingly, patients had CPIs with 12 unique individuals each day (half were other patients). Overall, patients spent half of their day (12.2 hours) in contact with another person.

A few thoughts. Studying social interactions and transmission in long-term care settings using these methods is quite brilliant. Given that LTCF cohorts are more stable with longer lengths of stay, it is easier to catch acquisition events through repeated screening. It is also important to note the huge amount of contact that LTCF residents have directly with each other - around 10 hours/day. This sort of social interaction among residents is not seen in acute care hospitals and goes a long way to explain why infection control in LTCF is so critical (and so extremely difficult). Given the richness of the CPI data, including frequency and duration of contacts, new mathematical models using these parameters could provide more accurate estimates of S. aureus transmission and effectiveness of candidate control strategies.

One issue that I think the authors might want to address in future studies is the use of spa typing to link transmission events. It is true that they also used susceptibility data, but it is my understanding (from one of Dan's earlier posts) that the discriminatory power of spa typing may be suboptimal. I hope Dan will provide his thoughts on this in the comments.

Additional reference: NPR Shots blog by Scott Hensley with comments from David Hartley on this study. 

Contact networks and infection prevention

Most pathogen transmission in healthcare facilities occurs via close contact, whether between healthcare personnel (HCP) or between HCP and patients. It is common sense that when it comes to transmission risk, all HCP are not equal. Consider the ICU nurse assigned to two patients during his shift, rarely venturing away from these two bedsides, versus the ID fellow or the respiratory therapist who contacts many patients across several units. Yet many models of disease transmission either don’t take contact network epidemiology into account, or don’t have sufficient data to understand the properties of HCP contact networks.

Our colleague Phil Polgreen and his collaborators in Iowa’s computational epidemiology group have constructed HCP contact networks using electronic medical record logins, validated the data using wireless sensors in our MICU, and applied the data to model the impact of various strategies to vaccination that focus on random application versus applying the intervention to HCP based upon degree (number) of contacts or distance (mobility) in the hospital. The figure below, from their recently published PLoS One paper, demonstrates the impact in a particular contact network (a) of vaccinating randomly (b), versus based upon degree (c) or distance (d).

This work has important implications for infection prevention practice. Imagine a year in which influenza vaccination is in short supply—see below for the impact on the disease attack rates in hypothetical scenarios where vaccination is based upon degree of contacts or mobility (distance) versus random allocation, in a sparse (a) or dense (b) contact network.

New tools for shoe-leather epidemiologists

Here's an interesting article in Salon on the use of social media to identify people at risk of STDs. The article contains an interview with Peter Leone, an Infectious Diseases physician at UNC, who discusses exploring social networks to expand the circle of at-risk persons instead of just relying on conventional  methods for contact tracing.

Graphic: Salon

Connected

Eli recently posted a piece on a study by Nicholas Christakis and James Fowler that looked at the spread of influenza in social networks. I'm currently reading their book, Connected, which explores many issues regarding social networks. It's an interesting read, particularly the chapter on how these networks influence health, including the contagion of obesity and suicide, and the complexity of transmission of sexually transmitted diseases. They also note how social networks could be exploited to improve health. Here's one interesting way: it's possible to achieve the effect of randomly vaccinating 99% of individuals in a population by vaccinating only 30% of the acquaintances of randomly selected individuals. There's also a TED talk on social networks by Christakis here. Lots of food for thought.

Why your friends spread influenza and you don't

There was a really interesting study by Nicholas Christakis and James Fowler in the May 15th Economist (here) and posted online (full manuscript here). The general thought behind the study was that people in the center of social networks, the ones with more friends or connections, would be more likely to be infected with influenza sooner. Thus, if you could identify people in the center they could serve as an early warning system for flu. The problem is, this would take a significant amount of effort. The insight into this problem comes in the form of what is called the friendship paradox. This 'paradox' suggests that your friends have more friends than you do.

To test this theory, Christakis and Fowler identified 319 students and then 425 of their friends. If friendship paradox would work in identifying influenza then the 425 should get flu earlier. So they followed the 744 students from September to December 2009 during the H1N1 epidemic and found that the friends, the 425 more likely to be in the center of the network, developed influenza (self-diagnosed and confirmed) around 2 weeks earlier. In fact, self-reported symptoms peaked 83 days earlier and visits to healthcare facilities peaked 46 days earlier in the connected group. Maybe this is a new method that could be added to other monitoring systems. Either way, come influenza season, stay away from your friends, especially the ones who are really friendly.