Want to stay alive? Wear a PFD / Lifejacket.

Wear a lifejacket (PFD). It’s a marine safety mantra. Yet people still opt out. A study of six years of RNLI incident data shows that in all inshore, coastal and near water activity wearing a lifejacket is indisputably the dominant factor in survival when things go wrong.

The simple, easy, but much neglected practice of wearing a lifejacket had the greatest impact on reducing casualty death in ‘Life at Risk’ incidents attended by the RNLI. according to an analysis of six years of lifeboat service data.

Incident survivability depends on lifejacket wearing

The effect of lifejacket was markedly visible in incidents where a person was in the water for a prolonged period, after for example, having fallen overboard, or being washed into the sea while fishing. The proportion of survivors wearing a lifejacket increasing linearly from about 35% at 10 minutes to 80% at 120 minutes. These results, directly from RNLI rescue statistics, form a crude composite survival index of direct and particular relevance to coastal and sea side marine users in the UK. Essentially if 100 people fall into the water, only 20 will survive 2 hours immersion without a lifejacket.

Brow to below decks: wear floation

These data make a case for wearing personal flotation religiously at all times: from the top of the brow until below decks. I argue this because of the high fatality rate among anglers and other coast-side users falling unexpectedly into the sea together the equally high risk in Norwegian data of fatality from unexpected falls between jetty and vessel.

Overall effect of lifejacket wearing

Just under half of people (42.4%) recovered by the RNLI across the whole variety of life at risk services wore lifejackets. This is a complex figure to understand, since a not everyone needing a RNLI rescue can be expected to wear lifejackets: for example at one extreme people seeking to self harm are unlikely to do so, while at the other neither are motorists and walkers being washed into the sea. Somewhere in the middle are SCUBA divers and open water swimmers. However recovery statistics tell their own story. Of those 42.4% of casualties who were wearing lifejackets, 94% were retrieved alive. Of the remainder, those not wearing lifejackets the survival rate fell to 73.1%.

The new knowledge this paper contributes is the brutally large effect on increased fatality among non lifejacket wearers who find themselves in hazardous situations among shoreline, coastal and near offshore marine users to whom the RNLI responded. This learning is particularly valuable because the data is so precisely fitted to marine users in the UK, and so is definitely applicable by yacht clubs, training providers and individual boat crews.

 Read the research for yourself

The research paper this blog is based on 1is nicely written. The results and discussion are entirely comprehensible even if you have have to take the minimal maths on trust. I am grateful to the authors for the public pre-print. I’d encourage you to have a read, because many of the secondary findings from this research have direct applicability too.

Study details: smooth, becoming a little geeky later…

Six years of Return of Service (RoS) data, filed by each lifeboat on completion of each ‘service’ or callout, were filtered to exclude incidents where neither local lifeboat crew nor national reviewers believed that there was ‘Life at Risk’ (LaR). 2094 services ‘Life at Risk’ services remained. To accommodate the needs of the regression model which was designed to test casualty survival, each Life at Risk Return of Service was disaggregated to 3119 casualty level data records in a way which preserved information about studied variables.

It would be entirely inappropriate to conduct a formal controlled trial where the outcome was death by drowning. So researchers are left to work with naturalistic data often collected for other purposes. Data like these are often maintained in the form of counts, which traditionally are treated with summary statistics and not much more. This leads to limited learning potential. These researchers have however managed to do much more because the dependent variable in this study ‘not surviving’ is (fortunately) a ‘statistically rare’ event in the context of their subset of Return of Service reports where Life was at risk.

The fact that not surviving was ‘statistically rare’ permits the use of a statistical tool: a simple stepwise log-linear (Poisson) regression model of casualty survival in RNLI rescues over six years which were judged to fall in the category ‘Life at Risk’ (LaR) service. The model dependent variable was the occurrence of a fatality, with 1 being equal to loss of life. In addition to lifejacket wear the other variables were:

  • seastate,
  • visibility at incident,
  • time to reach casualty from RNLI reciept of information,
  • windspeed at incident,
  • whether the service was conducted at night,
  • swelll height
  • whether the person was immersed in the water
  • sea surface temperature
  • which of four categories of service: commercial, leisure, person, other) the report fell into. Casualties reported as ‘other’ are atypical, e.g. involving aircraft or motor vehicles.

Of these variables, the Sea Surface Temperature (SST) is the lowest quality, being imputed to casualty record data from the nearest Sea Surface Temperature recording station at the relevant time. This was necessary because the Return of Service reports do not routinely record SST.

A regression model allows experiments to ‘test’ the effect of every variable. In this naturalistic experiment it’s difficult to determine cause with absolute certainty, but the authors have been reasonably conservative and the data has not been over interpreted. Usual steps to preserve model integrity have been taken, for example the ‘commercial’ category was dropped because of co-linearity with other categories.

  1. Pitman, S. J., Wright, M., & Hocken, R. (2019). An analysis of lifejacket wear, environmental factors, and casualty activity on marine accident fatality rates. Safety Science, 111, 234–242. https://doi.org/10.1016/j.ssci.2018.07.016 ↩︎