What data source is used?

The content from this web site is derived from the following data sources:

  • The publicly available Adverse Event Reporting System (AERS) Data from the U.S. Food and Drug Administration. This database contains the safety reports sent to the U.S. FDA since 2004.
  • The U.S. National Library of Medicine UMLS RxNorm vocabulary for drug names.
  • The U.S Food and Drug Administration Structured Product Labelings.
  • The MedDRA terminology for safety event reporting.
  • Nhumi Technologies internal anatomical terminologies and anatomical models.

Nhumi puts its best effort to import and update the data sources regularly. The statistics can therefore change over time as new reports and data are imported.

How is drug safety assessed?

When a patient takes a U.S. marketed drug and suffers from a serious adverse event at the same time, a report is sent to the FDA with a set of information describing when the event occured and how related it seems to be to the drug. The set of collected reports is consolidated by the U.S. FDA into the Adverse Event Reporting System database and is monitored by experts who analyze whether an adverse event is potentially caused by the drug or not. To assess this causal link, a comparison is made between the drug and all other drugs: if the adverse event is more specific to this drug than to any other drugs, then it is potentially linked to this drug. The PRR value reported on several tables of the Nhumi Drug Safety web site measures how specific the adverse event is: the higher the PRR, the more specific to the drug.
The safety of a drug is therefore assessed by a statistics that compares one drug to all others. Sometimes, such comparison can be unfair as all the other drugs do not treat the same disease. To address that issue, Nhumi Drug Safety web site makes it possible to compare one drug to another one or to a group of other drugs (click on the "Change" link in the "Comparing Drug To" box). It is also possible to select only the cases or reports that correspond to the same prior medical condition: this drug will then be compared to other drugs that have been used when patients had the same prior condition.

Such analysis does not provide information about how frequent the suspected adverse reaction can occur but it tells whether a reaction is potentially due to the drug.

What is a case?

A case is a report of a safety issue related to the drug of interest. Nhumi performs several operations to clean, normalize and aggregate the cases coming from the U.S. FDA AERS data sources. The number of cases is furthermore aggregated over several packagings for a single drug name (brand name).

How does Nhumi process the AERS data?

The Adverse Event Reporting System (AERS) Data from the U.S. Food and Drug Administration consists of independently submitted reports that are provided as-is. Therefore, the naming is not always consistent and reports are given in chronological order. To provide as much insight into the data as possible, Nhumi processes the AERS data before loading it into the Drug Safety website.

  • Reports that were not confirmed by the drug's manufacturer are filtered out.
  • If a report has one or more followup reports, only the latest report is considered.
  • Drug names are harmonized by simplifying the AERS names and matching them against the U.S. National Library of Medicine UMLS RxNorm vocabulary for drug names.
  • Case reports are grouped by drug.
  • If the resolved RxNorm entry is a brand name (BN) or ingredient (IN), the case is added to that brand or ingredient's data. If it resolves to another RxNorm term type, we find the most related brand (for branded entries) or ingredient (for unbranded entries).
  • Several other cleanup steps ensure consistent data. For instance, the outcome (OT) field in AERS is interpreted to detect the seriousness of the reported adverse event.

What is a body map?

A body map represents medical conditions on a virtual model of the human anatomy highlighting the parts that the medical conditions relate to. The map between a medical condition and an anatomical structure is described in the section "What is the map from a condition to the human body?". The way the body parts are highlighted is explained in the section "What does the color on the body map mean?"

What is the map from a condition to the human body?

The map is built to link events, conditions or procedures to the body parts or systems (called sites below) that can be directly identified with the anatomical root causes of the events, conditions or procedures. The purpose is to identify the anatomical structures that are the locations of the biological processes responsible for an adverse events, a medical condition or a procedure. It is very similar to the pathophysiology of the medical condition but is also defined for procedures or general events.

The map is computed differently based on the type of information:

A) For adverse events, medical conditions and results

The finding sites are defined the same way for adverse events, medical conditions or test results. They are calculated according to the following procedure:

  • A.1. When the condition has a well identified root cause which is directly associated with a single anatomical structure, the finding site is defined as this structure.
  • A.2. If A.1. does not apply and the condition is directly associated with a single anatomical structure, the finding site is defined as this anatomical structure. Such case corresponds to the condition being its own cause.
  • A.3. If A.1. and A.2. do not apply, the site is undefined.

A finding site can therefore be interpreted as a well identified location of the biological process causing the event, condition or test result to occur.
When many different mechanisms can occur to explain an event, condition or procedure, the site is set to "Body".
When the condition is not anatomical such as "Family history of Alcohol", the site can be set to undefined.

B) For actions or intervention made on a subject

The procedure sites relate to procedures, that is, any action or intervention made on a subject. They are defined as:

  • B.1. When the procedure has a well identified root cause or intent which is directly associated with a single anatomical structure, the procedure site is defined as this structure.
  • B.2. If B.1. does not apply, the site is undefined. That is, even when the procedure is directly associated with a single anatomical structure but the latter is not the intent nor the reason for the procedure, the site is undefined

A procedure site can therefore be interpreted as the location of the biological process that was targeted by the procedure.


For A.1:

  • "Insulin dependent diabetic" has a direct cause "destruction of insulin-producing beta cells of the pancreas" which is associated uniquely to the anatomical structure "Pancreas".

For A.2:

  • The finding site of "Injury of the arm" is the arm.
  • "Pain" which can be found in all body parts is associated to the finding site "body" that includes most anatomical structures.

For B.1:

  • "Thrombosis prophylaxis" has a direct intent: reduction or prevention of "Thrombosis" which is associated with the anatomical structure "Veins".
  • A drug injection site would not be considered but rather the anatomical structure that is associated with the injected compound.

For B.2:

  • "Injection of a product in arm" without stating which product was injected is not mapped.

What does the color on the body map mean?

The colors indicate the most relevant anatomical structures according to a scale that is shown on the right side of the body: the most relevant structures are shown in red and the least relevant in light blue. The set of anatomical structures shown on the body depends on the view and the zoom level. The following paragraphs explain how the highlight is calculated when the body map relates to adverse events. The term "finding site" is used to denote the anatomical structure or body part that the adverse event map to (according to the map described in the section "What is the map from a condition to the human body?":

  1. The regional view aggregates adverse events according to the regions of the body. The aggregation is defined by the zoom level. At start, only large regions of the body are considered. Adverse events are therefore aggregated on the head, neck, legs, etc. When zooming in, the regions are split into sub regions and the adverse events get aggregated more precisely based on their finding sites. Adverse events not only contribute to the score of their finding sites but also to all other regions containing their sites. The score of the head corresponds therefore to all adverse events whose finding sites include a sub part of the head. On the other hand, an adverse event does not contribute to the regions included into its finding site. If a finding site relates to a system for instance (e.g. cardiovascular system) and covers the whole body, no specific region is associated and the adverse event will be considered as "non regional". A system such as the visual system will however still be associated with adverse events of the head region.
  2. The systemic view aggregates adverse events according to the system they relate to. A system is defined by a process and can be as generic as the venous system or as precise as metabolic pathways. The aggregation is defined by the zoom level but changes less than the regional view as most systems cover a large part of the human body. Adverse events whose finding sites are included into a system will be associated to that system. Such view enables the user to see anatomical structures that cover several body parts and are not regional in nature.

What does PRR (shrinked) and PRR mean?

PRR is computed per adverse event and stands for the Proportional Reporting Ratio: it is computed as the fraction of reported adverse events for a given drug divided by the fraction of reported adverse events for all other drugs. The larger the PRR the more likely the association between the adverse event and the take of the drug. Please note that such association is hypothetical only and needs to be confirmed by a confirmatory experiment. A value close to 1 means that the adverse event does not a priori have a link with the drug.

PRR (shrinked) is identical to PRR except that it is smoothed when the number of observations is small. This smoothing procedure relies on a posterior estimate of PRR using a non uniform prior that favors smaller fraction and disappears when the number of cases gets in the 150 range. The posterior distribution over PRR is used to compute a lower bound on PRR at a 95% confidence interval. This lower bound is reported here. The same values are computed for adverse events, prior conditions or anatomical structures. In each case, the higher the number the more relevant or specific to the drug is the adverse event, medical condition or anatomical structure. The fractions used to compute the PRR (shrinked) are derived using the "Do calculus" to ensure that confounding factors and biases are reduced to a minimum. Such fraction is computed by conditioning on age and gender and by assuming that taking the drug is an intervention and not an observation. To further remove any bias, it is advised to select a priori condition and/or a co-medication that could explain the link between the adverse event and the drug under investigation.

What values of PRR are reported?

PRR greater than 2 are reported. Values lower than 2 are considered as irrelevant to indicate due to the level of noise in the data.

What is the Nhumi Drug Safety Scale?

The Nhumi Drug Safety Scale is a measure of safety using observed data only. It is computed by looking at the likelihood of getting a serious adverse event when taking the drug. The biases due to age, gender and medical conditions when taking the drug have been factored out. This scale is only an indication to facilitate the navigation on the website. It is not an exact measure as it is automatically computed from noisy data but is meant to help the visitor to explore alternative drugs with a different safety profile.

How can the number of cases be interpreted?

The number of cases represent the number of reported adverse events that are suspected to be related to the drug. This number does not reflect the whole set of adverse events that should be observed in the population nor does it model the fraction of patients who had an adverse event when taking the drug. The information carried by this number is therefore limited and should be interpreted with caution. It offers an indication though about which conditions have been observed in subjects taking the drug.

Filtering on the reason for the drug can provide a better information by removing the bias in the selection of the reports. It is therefore advised to always compare safety information by filtering on the medical conditions first.

How can the heat map be interpreted?

A red spot on the human body indicates the anatomical structures most likely to be adversely affected by the drug if any adverse event might occur. Please note that this is only a relative indication and should be interpreted with caution as the number of cases does not reflect the whole set of adverse events observed in the patients who took the drug.

How to contact Nhumi Technologies?

Send us an email at info@nhumi.com