FDA Repairs its MAUDE Database

I have posted several times this year about an error in the FDA MAUDE database (see here, here and here for previous posts).

To summarize, the FDA's MAUDE database captures reports of malfunctions and injuries associated with medical devices. The MAUDE database is supposed to capture patient outcome information like:

• Did the patient die during/as a result of the device malfunction?
• Was the patient hospitalized due to the malfunction?
• Was the patient disabled?
• etc.

About 2 years ago, the FDA's MAUDE database stopped reporting patient outcomes and replaced these outcomes with a cryptic code of "8.".

As you might imagine, this is a pretty serious omission. It seems kind of important to know what happened to the patients when a device malfunctions.

I contacted the FDA earlier this year and informed them of the error.

Today, I received an email indicating that the database has been fixed and the outcomes are back!

FDA was also kind enough to explain what "8." meant:

End of Story.

Update # 2 on "8." MAUDE Database

I've previously written (here and here) about an error in the FDA medical device adverse events database known as MAUDE. The error prevents the database from reporting patient outcomes (death, hospitalization, etc.).

I received correspondence from someone at the FDA last week indicating that they may have found a fix (see screenshot of email).

However, so far, they have not implemented the fix. You can see this in two ways:

First, if you download the patient outcome file from the FDA website and open it up, you see a series of "8"s where the outcome data should be (see pic).

Second, you can query the MAUDE database using the OpenFDA platform and count the distribution of patient outcomes in the year:

As of this moment, if you click on this link, you will see the distribution of patient outcomes in the MAUDE database from the beginning of 2016 until yesterday. You can see from the pic below that there are no outcomes, only variations on "8."

Update on "8." and MAUDE Patient Outcomes

I have previously posted about the FDA's adverse events database for medical devices (known as MAUDE). In past years, the MAUDE database had captured patient outcome information (whether the patient died, was hospitalized, suffered some form of disability, etc.).

The database no longer reports patient outcomes (it stopped doing so in late 2015). Instead, the database field that should contain outcome data contains the cryptic entry of "8." (the number 8 with a period). I recently contacted the FDA, asking why MAUDE no longer provided patient outcome data. I received the following responses (thus far). I've edited out identifying information (and superfluous text).

This is from an email dated 28 July 2017. It's moderately encouraging. In particular, it's encouraging to know that the patient outcome information  is _still_ captured, and it will probably be added back to the database once they determine what the error is.

I will post updates as they develop.

 Dear Dr. Danese,

Thank you for contacting the Division of Industry and Consumer Education (DICE) at FDA's Center for Devices and Radiological Health (CDRH) xxxxxxx@fda.hhs.gov e-mail account. As far as I am aware, patient outcome information is still collected on so adverse event reporting forms. So, at this time I am not clear on why this information is not included in the data sets for 2015 and beyond. I am in the process of researching the issue and contacting the medical device reporting staff. I apologize for the delay but hope to have an answer for you next week. .... Sincerely, xxxxxxxxxxxxxxxxx xxxxxxxxxxxxxxxxxxx Division of Industry and Consumer Education Office of Communication and Education Center for Devices and Radiological Health U.S. Food and Drug Administration

I then received a follow-up email on 4 August (Friday)

Dear Dr. Danese, 

Thank you for your patience while I looked into your inquiry. I have spoken with a member of the Information Analysis Branch and it seems some of the patient outcome information may have been inadvertently excluded from the downloadable zip files. xxxxxx has kindly agreed to look into this issue to figure out what happened and how the problem may be corrected. Please feel free to contact xxxx directly. xxxxx name is xxxxxxxxxxxxxx and xxxxx email address is xxxxxxxx@fda.hhs.gov. I hope you have a pleasant weekend. Sincerely, 

xxxxxxxxxxxxxxxxxxxxxxxxxxx 

xxxxxxxxxxxxxxxxxxxxxxxxxx

Division of Industry and Consumer Education Office of Communication and Education Center for Devices and Radiological Health U.S. Food and Drug Administration

And then this email from the Information Analysis Branch (dated 7 August)

Paul, I was notified of the issue by xxxxxx on Friday, and we are currently investigating. We will get back to you when we know more. Thank you. xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx Information Analysis Branch Division of Post Market Surveillance Office of Surveillance and Biometrics Center for Devices and Radiological HealthUS Food & Drug Administration 

Clustering of Drugs by Side Effect Profile

Others (here, here, here + others) have clustered medications based on side effect profiles. Everyone has done it a little differently.

In this post, I perform a similar type of clustering using slightly different methods (details forthcoming). I'm not providing m̶u̶c̶h̶ any of the methodological details in this post, but I just wanted to show a pretty plot with nice clusters.

Drugs were clustered using t-SNE (here and here), which is a method that is remarkably good at taking high-dimensional data (in this case, thousands of different side effects) and clustering it into 2-dimensions while maintaining much of the "nearest neighbor" information that was present at high dimensions.

You can see the 2-D "cluster plot" (after t-SNE transformation) below.

• Hover over the data points to see drug names.
• I colored the data points by running k-means on the 2-D data (very quick and dirty and probably not the best approach).
• Still, you can see pretty good functional clustering. For example: 
• the dark green points on the left edge of the plot are almost all related to cholesterol and lipid pharmacotherapy. If you zoom in, you will also find daptomycin (an anti-bacterial), but it clusters there b/c it is associated with reports of myopathy and rhabdomyolysis (adverse events that are somewhat characteristic of the "statins").
• red and blue points on the left edge are primarily involved w/ Type 2 diabetes.

Trump submitting adverse event reports?

Found this "classic" adverse event report submitted to the FDA in 1994. 76 year old male patient taking Sinemet and Xanax with a report of the following issues:

• asthenia
• coordination abnormal
• dysphagia
• med error

The kicker is in the comments section: 'Pharmacist "jokingly" told patient to take the medication with Vaseline, [which] the patient did.'

• First, I'm not sure why jokingly is in quotes. Was the reporter quoting the pharmacist, in which case the quotation marks emphasize a literal and sincere description of the communication between pharmacist and patient? Or are the quotation marks meant in the ironic-2017-Sean-Spicer-fake-news sense, in which case the pharmacist comes across as malevolent? Tough call.

• Second, the report labels the Sinemet as the suspect medication! I'm going to go out on a limb and speculate that the Vaseline itself was causing the dysphagia, but don't quote me on that.

I'm trying to make a dictionary of drugs named in FAERS and their active ingredient(s).

Let's see how many ways we can abbreviate, misspell name a drug that contains paracetamol/acetaminophen:

1. acamol
2. aceminophen
3. acetam
4. acetametamin
5. acetamilnophen
6. acetamin
7. acetaminaphine
8. acetamino
9. acetaminofen
10. acetaminohen
11. acetaminop
12. acetaminopehn
13. acetaminopen
14. acetaminoph
15. acetaminopham
16. acetaminophe
17. acetaminophen
18. acetaminophen0
19. acetaminophin
20. acetaminophn
21. acetaminphen
22. acetamiophen
23. acetamonophen
24. aceteminophen
25. acetminophen
26. acetomeniphen
27. acetominopehn
28. acetominophen
29. actigrip
30. actyaminphen
31. adetaminophen
32. algisedal
33. algotropyl
34. alpiny
35. alvedon
36. alvedone
37. amidrine
38. anhiba
39. antalvic
40. apap
41. aplexil
42. aracetamol
43. atasol
44. butal cf acetamn
45. calonal
46. calpol
47. car panadol
48. claradol
49. cocodamol
50. codamol
51. codoliprane
52. coltalin
53. combiflam
54. comtrex
55. dafagan
56. dafalgan
57. daflagan
58. defalgan
59. depalgos
60. depon
61. dexamol
62. dextr neo citran
63. di antalvic
64. dolipran
65. doliprane
66. dristan
67. efferalgan
68. endocet
69. esgic
70. excedrin
71. excedrin migraine
72. exedrin
73. exedrin asa
74. fervex
75. feverall
76. fioricet
77. frenadol
78. grippostad
79. hycet
80. hydroc/apap
81. hydroco apap
82. hydroco/acetaminophen
83. hydroco/apap
84. hydrocod/acetam
85. hydroodone/apap
86. ixprim
87. kolibri
88. lamaline
89. liq tyelnol
90. loratab
91. lorcet
92. lortab
93. mapap
94. maxidone
95. maxiumum midol pms
96. midol
97. midol pms
98. midrin
99. migrazone
100. miradol
101. mission supac
102. neo citran
103. neocibalena
104. norco
105. ofirmev
106. osteo panadol
107. oxy/apap
108. oxycod/apap
109. pactiv
110. panadeine
111. panadol
112. panodil
113. parace
114. paracematol
115. paracemtamol
116. paracet
117. paraceta
118. paracetam
119. paracetamil
120. paracetamo
121. paracetamol
122. paracetamol0
123. paracetamolo
124. paracetmol
125. paracetmol aporex
126. paracetramol
127. paractol
128. parad
129. paralgin
130. paralyoc
131. paramol
132. percocet
133. perfalan
134. perfalgan
135. pro dafagan
136. prontalgine
137. proparacetamol
138. pyrinazin
139. roxicet
140. saridon
141. sinex
142. sinutab
143. tachidol
144. tachipirina
145. traiminic
146. tramcet
147. trimanic
148. tyelnol
149. tyenol
150. tylenol
151. tylenolpm
152. tylonal
153. ultracet
154. ultraset
155. xartemis
156. xolox
157. zapain
158. zydone

FAERS Popcorn

Gotta love some of the things you find in FAERS, including the deadly popcorn as a secondary suspect for confusion sedation, and vasodilation.

FDA MAUDE, Patient Outcomes and "8."

The FDA's MAUDE database contains reports on malfunction of medical devices. You can run more comprehensive searches here.

There's a lot of information buried in the MAUDE database, including the outcomes of patients who were on the receiving end of a malfunctioning device.

Specifically, the MAUDE database uses a 1-letter code to list the following possible outcomes:

• Congenital Anomaly (C)
• Death (D)
• Disability (S)
• Hospitalization (H)
• Life Threatening (L)
• Other (O)
• Required Intervention (R)
• plus a few others that are not particularly interesting.

Lately, the FDA has stopped listing patient outcomes and have replaced them with the following cryptic message: 8.  That's right, the digit 8 and a period. I have no idea what that means, but there has been no patient outcome information (other than "8.") in the MAUDE database for ALL of 2016.

Let's take a look at the rise of "8." (see blue line below).  If we go back to 2012 and 2013, we can see that only 1 - 5% of MAUDE reports had an outcome of "8."

Things start to change in the spring of 2015, when "8." begins an exponential march upward.

By October 2015, 100% of the MAUDE reports show an outcome of "8." and it hasn't relinquished its stranglehold since.

I have no idea what "8." means. 

• Is it a database error? 
• Does it mean that the FDA has stopped releasing this information?
• Something else entirely?

If anyone knows what "8." means, please let me know!!!

Here's some additional weirdness.

The graph shown above only goes back to late 2012.
You can go all the way back to 1991 and plot MAUDE outcome data.
What you see are periodic spikes of "8." (see blue line in chart below).

Again, I have no idea why "8." would take up 30+% of the patient outcomes in September of 2004, but it's there.

We need more of this.

And then there's MAUDE

FDAble also provides clients with adverse event and malfunction information for medical devices using the FDA's MAUDE database. MAUDE is an acronym for Manufacturer and User Facility Device Experience Database. I think we can all agree that this is a horrible name, unless you are a fan of Bea Arthur

.

The MAUDE database is weird in so many ways (more about that in a future post), but one of its remarkably nice features is that it provides a free-form narrative describing the problem associated with each report. In contrast to MAUDE, the FDA does not provide the narratives for FAERS data (unless you file a FOIA request and are willing to wait about 1 month).

From the looks of it, the FDA currently receives about 80-90 thousand MAUDE reports per month (which is roughly the same magnitude as the number of FAERS reports received per month as well). You can also see that MAUDE recently celebrated its 300th month anniversary (starting in December of 1991).

The FDA started including manufacturer reports in August of 1996, which explains the giant leap in the # of reports at that time.

FAERS Reports on Counterfeit Medications

AARP magazine is using some FAERS data that I provided as part of a forthcoming article on adverse events and counterfeit medications.

They were interested in FAERS reports in which counterfeit was part of the MedDRA preferred term. This included:

• suspected counterfeit product
• product counterfeit
• counterfeit drug administered

A common co-reported adverse reaction was drug ineffective (not surprising).



The patient outcomes in this group of cases is disturbing, with 703 cases reporting death.

In contrast to the overall FAERS database, the reports concerning counterfeit drugs are dominated by males.

There is also a spike in the # of reports of counterfeit products in 2015. Keep in mind that the FDA has not yet released data for the 4th quarter of 2016, but it seems unlikely that one additional quarter's worth of data will push 2016 above the 2015 number.

Here's the distribution of "occurring countries." More than I expected in China since this database is focused on the USA.

Age / Gender Distribution in FAERS Reports

If you break FAERS reports by gender / sex, you'll see that the number of reports for females is remarkably larger than the number of reports for males. My speculation is that this has to do with the notion that men feel that filing a report complaining about their meds makes people question their masculinity. Again, no idea if that's true, but it seems reasonable.

Below, you can see histograms of FAERS reports partitioned by patient age and gender.

Some highlights:

1. The female age distribution is in pink. 
2. The median female age is 55 years.
3. The male age distribution is in blue (most of the male distribution shows up as purple because it is overlaid on the pink female distribution (pink + blue = purple). 
4. The median male age is 58 years.
5. You can see the # of female reports dominates over the male reports.
6. There is one exception and that is with young patients (reports for males are more common for newborns up to about 13 yo).
7. Then the shape of the female and male distributions is roughly identical except for the "hump" from ages 17 to 45 in the female distribution, which means that females report more frequently in all age groups from 17 years and up, but they report much more frequently from 17 to 45ish.

More importantly, my son's moonshot has gotten better.

Improved FAERS Reports

I spent some free time in the past month or so improving the summary documents that go with the FAERS reports that we sell (click here to download a sample report).

The pics below are for FAERS cases in which OxyContin is a primary suspect.

Here is a chart of new OxyContin cases per year. Clients ask for this a lot, and I've finally automated the process of generating this type of chart.

This bar-graph shows OxyContin adverse events grouped by MedDRA System-Organ Class (SOC). This is also something that many clients have asked for in the past. I also have automated grouping of adverse events by high-level group term (HLGT), and high-level terms (HLT), which I think may be helpful for regulatory filings.

I'm not sure how useful this heat-map is, but it looks nice. The title is self-explanatory.

Finally, here is an awful photo of the moon that I am proud of since my 10-year old son took it in the freezing cold about 1 week ago. He got a telescope for Christmas and he held my phone up to the eyepiece with his shaky hands. I think you can actually see a crater or two in the photo. The actual view through the eyepiece with our human eyes was far more impressive. 

# of FAERS Cases w/ non-serious outcomes has risen?

Okay. I think this is slightly more interesting than the previous post.

Here's some background on the FDA Adverse Event Reporting System (FAERS)

Every AERS/FAERS cases has the possibility of listing a serious outcome from an adverse event.

These outcomes fall into the following (non-mutually exclusive) categories:

• congenital anomaly
• death
• disability
• hospitalization
• life-threatening
• required intervention to prevent permanent impairment
• other (serious outcome)

If you plot the number of serious outcomes reported per quarter versus the number of reports per quarter, they move almost in lock-step until around late 2010 or 2011 (compare red and blue lines below).

After late 2010, the # of serious outcomes begins to lag the # of reports, which indicates one of three things:

1. The increase in the # of FAERS reports submitted over the past six years is mostly due to an increase in submission of reports w/ non-serious outcomes.
2. The increase in the # of FAERS reports submitted over the past six year is mostly due to an increase in submission of reports w/ unknown outcomes.
3. Some combination of points 1 and 2.

Data and code are here: https://github.com/pdanese/faers_counts_Q3_2016

Below is a slightly different way to view the same information.
The plot shows the ratio of serious patient outcomes to the number of reports filed per quarter. You can see that in the "early 2000s" the number of serious outcomes per report was around 1, but it has dropped in recent years 0.6 to 0.8 range. So, about 20 -- 40% of the FAERS reports since late 2012 do not report a serious outcome.

FAERS Report Sources Almost Non-Existent These Days

Yesterday's post is admittedly boring. I mean, AERS / FAERS reports have been going up over time (for the most part) with occasional "pauses" but is there anything more interesting in this gross analysis of the FAERS database?

Mayyyyyybe.

First, another boring bit.

Each FAERS report received by the FDA usually lists multiple medications (not a big surprise--many people take multiple medications). Each medication listed in a FAERS report gets its own record. So, we can ask:

Has the # of reported medications per report changed over time?

Short answer: probably not.

Below is a chart of the # of drug entries per quarter (blue line).
You can see that it roughly correlates with the # of reports per quarterly release (red line).


You can be a little more rigorous and plot the mean number of drug entries per report (per quarter) [see chart below]. I don't want to do the statistics on this, but my suspicion is that the regression line would be relatively flat, which means the # of reported meds per report hasn't changed much over time.

It looks as though there are about 3.5 medications listed per FAERS report and that's held steady.

I guess that's good. Maybe that means that there hasn't been a shift in adverse events related to the sheer number of drugs taken (per patient) over the past 20 or so years. I know that's a really speculative conclusion, but I suspect there's a kernel of truth in there.



But there's more information included in FAERS reports!

For example, each report can indicate the source of the adverse information. This is formally known as the Report Source. Some possible report sources include:

• Health Professionals (physicians, etc.)
• Company Representatives (the pharmacovigilance department let's the FDA know)
• Scientific Literature (information is published in a medical journal)
• Pharmaceutical Distributor
• etc. etc.

The chart below shows the number of sources listed per FAERS report over time and it indicates that the report source information is essentially dead. . These days, less than 1 FAERS report out of 40 contains report source information. This doesn't seem like absolutely critical information for the FDA to publish, but I'm still surprised that it's fallen by the way-side.

You can download data and code here: https://github.com/pdanese/faers_counts_Q3_2016

New FAERS Data Q3-2016

FDA recently released FAERS data for 3rd quarter of 2016. Seems as though we've hit a flat period where the # of reports per quarter is moderately stable in the low 300,000-range. Maybe it's just a pause.

Putting the Band Back Together

Okay. I used to have a blog and I'm trying it again. Hopefully, this lasts longer than most New Year's Resolutions. This blog will deal with the FDA and health data. In particular, the FAERS and MAUDE databases and whatever else strikes my fancy.