Data backup plan: exact copies the office can restore
What gets copied, where the copy sits, and the last time a person opened a restored note.

A data backup plan is the procedure for making exact copies of electronic patient information and keeping those copies so they can be read back. The Security Rule requires that plan. It does not set how often you copy, or which product you use.
The backup job had been green for months. When the server failed, the only copy anyone could find was three weeks old, and the notes from the last two clinics were not in it. The green light had been copying an empty folder.
The requirement sits in 45 CFR 164.308(a)(7)(ii)(A). The data backup plan is a required implementation specification under the contingency plan: establish and implement procedures to create and maintain retrievable exact copies of electronic protected health information. HHS describes the Security Rule on its security rule overview. NIST’s guide for a small organization is NIST SP 800-66r2. “Retrievable” and “exact” are the words that do the work. A copy you cannot open, or a copy that is missing the last week, is not the copy the spec describes.
The broader downtime story, including who sees patients while the EHR is down, already lives on the contingency plan page. This page is only the copies: what is copied, where it sits, how often, who can get it back, and the last time someone proved a note was actually in it. Keep that procedure with the managed IT work so the job is a named task, not a light on a screen nobody opens.
What exact and retrievable mean
Exact means the copy can stand in for the original for the information you decided to protect. A backup of the EHR database is not exact if scanned documents live on a different disk that the job never touches. A backup of the documents is not exact if the schedule and the notes are elsewhere. Write the list of stores before you write the product name. The EHR, the document folder, the billing extracts you keep locally, and any device that holds the only copy of a download. If a store is not on the list, it is not backed up, and the page should say that in those words.
Retrievable means a named person can get a specific record back without the original system being healthy. If the only way to read the copy is the server that just failed, the copy is not retrievable. If the password for the backup is known only by a person who left, it is not retrievable. Write who can restore, where the second key is, and what “a specific record” means in practice: one patient’s note from a named date, not a mountain of files nobody can search. The encryption decision belongs here, because a scrambled copy with a lost key is a copy you cannot use.
Maintain means the copies keep existing. A job that ran once at go-live and was never checked is not a maintained plan. A drive that filled up in April and has been failing since is not a maintained plan. The procedure names how you know the job ran: a report a person reads, not only a green icon. Someone looks. The activity review can include one line for the backup report so it is not a separate religion that nobody attends.
The rule does not set a daily, hourly, or weekly cadence. You set it, from how much information you can stand to redo. A clinic that charts all day and would lose a day of notes if the copy is from yesterday morning should say so and copy more often. A cadence you will not check is worse than a slower one you do check. Write the gap you accept, in hours or in a clinic session, so the next manager does not discover the gap during the failure.
Where the copies live
A copy that sits on the same disk as the original dies with the original. The procedure says the copy is somewhere else: another device, another room, or a service whose failure is not the same failure as the EHR server. You do not have to buy a particular cloud. You do have to be able to point at the other place. If the “other place” is a drawer in the same office, say what happens if the office is the thing that is unavailable. The contingency page covers getting in the building. This page covers whether the copy was ever outside the machine that failed.
More than one copy is a choice, not a sentence in the rule. If you keep two, write what each one is for. One may be the recent copy you would restore this afternoon. One may be the older copy you would use if the recent copy was itself damaged or encrypted by the same event that hit the EHR. Say how old the older one is allowed to be. An older copy that is also three weeks stale does not answer a different question. It answers the same question twice.
Portable copies are devices. A drive in a bag follows the device and media rules when it is reused or thrown out, and the facility page if it sits in a room. Label it as a backup, not as a spare disk someone may format. The person who takes it off site is named. An unlabeled drive in a desk is not a location. It is a rumor.
Vendor-hosted EHRs still need a plan. “The vendor backs it up” is a start only if you know what they copy, how far back you can ask, how you request a restore, and how long that request takes. File the answer. If the contract is silent, the plan says silent, and names the date you will ask. A brochure is not a procedure. A named ticket path and a tested restore are.
The restore you have actually done
A backup you have never read back is a hope. Once a year, or after a change in the job, restore one known note to a place that is not the live chart, and compare it to the original. The integrity page uses that same restore as a check that a note was not destroyed. This page uses it as a check that the copy exists and opens. Write the date, what you restored, and whether it matched. If it did not, the plan is wrong until the job is fixed. Do not wait for a failure to learn the folder was empty.
Pick the note on purpose. A note from last week, a scanned document, and a schedule entry are three different stores. If you only restore the database and never a scan, you have not tested the scan store. Rotate what you pick. The test is small. It does not require a full disaster drill. It requires one record a person can recognize. If the restore needs the vendor and takes two weeks, write that delay on the plan, because that delay is part of what “retrievable” means for you.
Failed jobs are part of the procedure, not an embarrassment to hide. The report that says the job failed needs a person and a same-week fix, or a written reason the fix is waiting and the date it will be done. A month of failures with a green dashboard someone did not open is how the three-week gap happened. The owner of the plan is the person who reads the failure, not the person who installed the software two years ago.
Owners do not get a private unbacked laptop. If the only copy of a download sits on the owner’s machine, that machine is a store on the list, or the download is forbidden. A plan that covers the server and ignores the laptop the owner uses at home is a plan for a building you do not work in. Add the line. The same identity and termination rules apply when the person who knew the backup password leaves: the password changes the same day, and the backup person confirms they can still open a file.
What the office can say out loud
The card is short enough to read at the huddle. What is copied. How recent the copy is supposed to be. Who to call if the EHR will not open. Where that person looks first. New hires who will never restore anything still need to know that a local-only spreadsheet is not the backup. Otherwise they will build a second record the job does not see.
Qventive’s assessments at 200 Passaic Street in Hackensack ask for the store list, the age of the newest copy, and the date of the last restore that a person actually opened. If the newest copy’s age is unknown, the required spec is not in use. The same page sits with the security record so a new manager is not inheriting a password that lives in one head.
When you change EHRs, backup products, or the place scans are saved, rerun one restore before you retire the old job. Two jobs that both claim to be the backup, with nobody sure which is current, is not a maintained plan. Turn the old one off on a written date, after the new one has produced a note you opened. A parallel month is fine. A permanent mystery is not.
Once a year, read the store list with the person who charts and the person who files scans. Ask what they save that the job might miss. Add it, or forbid it, the same month. Walk one restore note with a new backup owner before they are the person you call on a bad morning. If they cannot find the key or the ticket path, fix the page before the outage, not during it.
Keep the restore notes and the failure notes for the same period as the other security records, and write that period on the folder. The backup spec does not set a retention number. A test thrown away after a month cannot show that the job was empty in March. Date every test. The history is how you know the green light means a note, not a folder.
Data backup plan: common questions
Is a data backup plan required?
Does the rule say the copy must be daily?
Is a green icon enough?
What if the EHR vendor says they back it up?
Does this page replace the contingency plan?
Let’s meet
A free practice technology assessment reads the store list, the age of the newest copy, and the last restore a person opened.