Six DisasterRecoveryOps Myths Health IT Teams Still Believe

Healthcare IT

Whenever a health-information system goes offline, a familiar set of reassurances starts circulating in the staff room: “we have backups”, “the vendor handles that”, “it is replicated, so we are fine”. Some of these are true as far as they go. Many fail the moment a real outage tests them. This article is about the administrative and IT side of healthcare services, not about clinical matters, and it uses the Ops Atlas’s treatment of DisasterRecoveryOps to sort six common beliefs into myth and reality. Nothing here is medical advice; it is a plain look at how technology teams think about recovery.

Myth 1: “We are spread over several zones, so we are covered”

Placing a service in more than one zone of a cloud region guards against losing one building or one rack, and that is sound design. But the Atlas entry on DisasterRecoveryOps is firm that this does not extend to losing an entire region, or the provider’s management plane for that region. The entry notes that for a single-datacentre problem, a simple restore from backup may even be enough, while regional events call for plans built to span regions.

Reality: know which kind of event your design survives, and write down the ones it does not.

Myth 2: “Our recovery target is the same as our average repair time”

These sound alike and are very different. A recovery time objective is a target you set in advance, a promise about how long a service may be out. Mean time to repair is a figure worked out afterwards from events that already happened. The Atlas lists the confusion explicitly among its myths. A team can have a respectable average and still miss its target on the day that matters, because averages hide the slow outliers.

Reality: set the target from the consequences of downtime, then measure drills against it.

Myth 3: “The provider runs the infrastructure, so recovery is their job”

A hosting provider keeps its own machines and networks alive. What happens to your particular records system, your configuration and your access arrangements is a separate question. The entry cites a major cloud vendor’s own documentation, which says plainly that recovery does not arrive by default with the platform. Neither the contract nor the dashboard restores your service on your behalf.

This point has regulatory weight too. According to the Atlas, the US HIPAA Security Rule requires a disaster recovery plan for covered systems, so responsibility for having a plan sits with the organisation, not just its supplier. For the exact obligations that apply to any given service, consult the organisation’s compliance team.

Reality: ask your provider precisely what they restore, how quickly and where, then compare the answer with your own target.

Myth 4: “Replication means our data is safe”

Copying data to a second location in real time is excellent for availability. Unfortunately, the Atlas points out, replication copies mistakes and corruption just as faithfully as good data. A bad update, an accidental deletion or a damaged file arrives at the second site within seconds. For that reason the entry recommends keeping older point-in-time copies in addition to live replicas.

The same cautionary lesson appears in a large 2024 incident described in the entry, where a faulty content update crashed millions of Windows machines. Switching to a standby helped little, because the fault itself was everywhere.

Reality: live copies and historical copies solve different problems; most services need both.

Myth 5: “Our data-loss target is a number of records”

When people discuss how much data they can afford to lose, they often say something like “no more than a hundred entries”. The Atlas, following NIST, explains that the recovery point objective identifies an instant instead of a quantity: the moment you are able to restore to, for example five minutes before the failure. That distinction matters because the volume of entries per minute changes through the day.

The entry adds another subtlety. A service rarely depends on a single store, so the worst delay across everything it uses decides the real exposure. If appointment data is copied every minute but the document archive only nightly, then the nightly delay is the figure that counts.

Reality: state the target as a time, then check every store against it.

Myth 6: “Once the cause is found, the service is back”

It is tempting to read an announcement that a root cause has been identified as the end of the story. The Atlas’s reading of a large cloud outage in October 2025 shows otherwise. The faulty name-lookup layer was pinned down about 50 minutes into the event, lookups worked again at the 2 hour 37 minute mark, ordinary launches of new virtual servers resumed at 14 hours 2 minutes, and the event was declared over at 14 hours 32 minutes. Diagnosis and restoration were separate milestones, many hours apart.

For any service that people rely on to book or reschedule appointments, that gap is the difference between “we know what is wrong” and “you can use it again”. Good communication uses the second definition.

Reality: define “restored” as something someone can observe, such as a completed test transaction.

A word on what a drill looks like

Several of these myths survive because nobody has run a realistic drill. A drill does not need to be dramatic. A team picks a quiet window, announces that the main system is “lost”, and times every step from the first alert to the moment a test transaction succeeds. They record who noticed, who decided, which credentials were needed and which steps surprised them. The value is in the surprises: an expired password, a missing permission, a document that points to a server that was retired.

The entry cautions that some practices should be avoided in the process, such as recovering by clicking through a management console from memory, or testing failover only on paper. A written, repeatable script beats improvisation every time.

A bonus myth about decisions

One more belief deserves a mention: that recovery is a purely technical race. The entry’s central teaching is that the minutes of deliberation sit inside the recovery budget. In the entry’s sample drill, noticing and deciding consumed 31 minutes while the systems took 33; the sum, 64 minutes, ran past the 60-minute goal. A decision deadline, computed ahead of time, is how teams stop the discussion from eating the budget. Even a small service benefits from naming who decides.

If you are curious how simple recurring routines help, our article on sharing household costs through a tracker with agreed rules shows the value of settling responsibility before it is tested, which is the same principle in another setting.

Where to go from here with DisasterRecoveryOps

None of these myths comes from carelessness. Each one is a reasonable simplification that works in ordinary times. The skill lies in recognising which simplification is carrying weight in your own setting, and in testing it before a bad day does. Our general piece on recognising when a common complaint needs professional attention makes a parallel point in a different field: knowing when to escalate is a decision best made ahead of time. For the technology side, the Atlas entry linked above is a thorough reference, and a calm read of its myth list is a sensible way in.