(87 days)
The intended use of the NHANES Reference Data Software Option for Hologic QDR Series X-Ray Bone Densitometers is to allow comparisons of a subject's estimated proximal femur (hip) Bone Mineral Density (BMD, in g/cm²) to reference data for sex and ethnicity matched controls. Published results from the NHANES study (A.C. Looker, et al., Osteoporosis International Vol. 5, pg 389- 409, 1995), including male and female proximal femur reference values for African American, Mexican American, and caucasian American subjects are included in this software option.
The NHANES Reference Data Software Option for Hologic QDR Series X-Ray Bone Densitometers provides proximal femur (hip) reference data for male and female Caucasian American, African American, and Mexican American subjects. The new reference data is used in conjunction with previously existing software on QDR Series densitometers, which allows comparisons of patient results to sex, ethnicity, and age matched reference values. The NHANES Reference Data Software Option provides 1) reference data for male and female African American and Mexican American subjects which was previously not available, and 2) the newest, most up to date male and female caucasian American reference data, while maintaining previous Hologic hip reference data and still allowing clinicians to generate and use local reference data.
The reference values contained in this software option were obtained from published results (A.C. Looker, et al., Osteoporosis International Vol. 5, pg 389- 409, 1995) of the National Health and Nutrition Education Survey III ("NHANES III") study, in which 7116 total subjects have been enrolled. Published reference values are stratified by sex, ethnicity, age (from 20 to 80 years), and analysis region (results are tabulated for the Neck, Trochanteric, Intertrochanteric, Ward's, and Total regions of the hip). Standard existing QDR densitometer software was designed to stratify reference values in exactly the same manner, and thus the published results are directly transferable to the Hologic QDR software.
Patient results for each analysis region can be compared to reference values both graphically and quantitatively. Graphical plots of the age dependent reference Bone Mineral Density (BMD) values corresponding to the sex and ethnicity of the patient are generated, with a marker placed at the position corresponding to the estimated patient BMD and age. Deviation "scores" are computed, quantifying (in population standard deviation units) the difference between the patient's estimated BMD and the mean value for either age matched (Z-score) or young adult (T-score) reference data. In addition, the patient BMD is expressed in percent of the age matched mean and in percent of the young adult mean.
The NHANES caucasian American female reference data differs slightly from the original Hologic data, due to differences in study design (inclusion/exclusion criteria, number of subjects, geographic region and methods of subject recruitment, etc.) and the unavoidable variability between different reference data sets obtained even under the EXACT same conditions. Thus both the Hologic reference data and the NHANES data are equally valid, and the physician may determine which data most accurately reflects the local population. Alternatively the software allows the physician to define and use unique local reference values of their choice.
Here's an analysis of the provided text regarding the NHANES Reference Data Software Option for Hologic QDR Series X-Ray Bone Densitometers, structured to address your specific points:
Since this device is a software option that provides reference data for bone densitometers, the concept of "acceptance criteria" and "device performance" as it typically applies to diagnostic algorithms (e.g., sensitivity, specificity) is not directly applicable in the same way. This submission is primarily about demonstrating that the new reference data is integrated correctly and provides the intended comparative functionality. The "performance" here refers to the software displaying the published reference data and calculating scores based on it.
Acceptance Criteria and Reported Device Performance
Given the nature of this software (providing reference data), the implied acceptance criteria revolve around the accurate integration and display of the published NHANES III data and the correct calculation of Z-scores and T-scores based on this data. The document implicitly supports these by stating the software was "designed to stratify reference values in exactly the same manner" as the published results and that it generates "deviation scores" correctly.
| Acceptance Criteria (Implied) | Reported Device Performance |
|---|---|
| Accurate integration of NHANES III reference data. | The reference values were obtained from published results (A.C. Looker, et al., Osteoporosis International Vol. 5, pg 389-409, 1995) of the NHANES III study. The software was "designed to stratify reference values in exactly the same manner" as the published results. |
| Correct stratification of reference values by sex, ethnicity, and age. | Published reference values are stratified by sex, ethnicity, age (from 20 to 80 years), and analysis region. Standard existing QDR densitometer software was designed to stratify reference values in exactly the same manner. |
| Ability to compare patient results to reference values graphically. | "Graphical plots of the age dependent reference Bone Mineral Density (BMD) values... are generated, with a marker placed at the position corresponding to the estimated patient BMD and age." |
| Ability to compare patient results to reference values quantitatively (Z-scores and T-scores). | "Deviation 'scores' are computed, quantifying (in population standard deviation units) the difference between the patient's estimated BMD and the mean value for either age matched (Z-score) or young adult (T-score) reference data." "In addition, the patient BMD is expressed in percent of the age matched mean and in percent of the young adult mean." |
Study Details
-
Sample size used for the test set and the data provenance:
- Sample Size for Test Set: Not explicitly mentioned as a separate "test set" in the context of validating the software's accuracy. The software integrates pre-existing published reference data. The reference data itself was derived from the NHANES III study, which involved 7116 total subjects.
- Data Provenance: The data comes from the National Health and Nutrition Education Survey III ("NHANES III") study. This is a prospective national survey conducted in the United States. The data covers male and female Caucasian American, African American, and Mexican American subjects.
-
Number of experts used to establish the ground truth for the test set and the qualifications of those experts:
- This question is not applicable in the traditional sense for this submission. The "ground truth" for the software is the published reference data itself. The "experts" who established this reference data were the researchers and statisticians of the NHANES III study, as cited in the publication by A.C. Looker, et al. Their qualifications would be those typical for epidemiologists, statisticians, and medical researchers involved in large-scale national health surveys. The document does not specify the exact number or detailed qualifications of these individuals but refers to their published work.
-
Adjudication method for the test set:
- Not applicable. The software is integrating published reference data. There is no diagnostic "decision" being made by the software that requires adjudication in the clinical sense. The Z-scores and T-scores are direct calculations based on the patient's BMD and the provided reference data.
-
If a multi-reader multi-case (MRMC) comparative effectiveness study was done, If so, what was the effect size of how much human readers improve with AI vs without AI assistance:
- No, an MRMC comparative effectiveness study was not done. This software provides reference data for clinicians to interpret existing DXA scans, not an AI-assisted diagnostic tool that would directly impact human reader performance in image interpretation. The software's role is to provide a statistical comparison context for the BMD values.
-
If a standalone (i.e. algorithm only without human-in-the-loop performance) was done:
- Yes, in a way. The "algorithm" here refers to the software's ability to accurately store, display, and calculate scores based on the NHANES III reference data. The document implies that the standard QDR densitometer software was already designed to handle such data stratification and calculations. The "standalone performance" ensures that the software correctly takes a patient's BMD, age, sex, and ethnicity, and then correctly retrieves the corresponding NHANES III reference mean and standard deviation to calculate Z-scores and T-scores as published. This functionality is independent of a human's input in the calculation process itself, though a human interprets the output.
-
The type of ground truth used (expert consensus, pathology, outcomes data, etc.):
- The "ground truth" for this software is the published statistical reference data from the NHANES III study. This data represents population-level bone mineral density distributions collected through a rigorous, large-scale epidemiological survey. It is not expert consensus on individual cases, pathology, or direct outcomes data in the usual clinical validation sense, but rather statistically derived normative values.
-
The sample size for the training set:
- Not applicable in the context of an AI/ML training set. The software does not use machine learning. It incorporates pre-computed, published reference data. The "training set" for the reference data itself was the 7116 subjects enrolled in the NHANES III study.
-
How the ground truth for the training set was established:
- Again, not a traditional "training set" for an AI model. The reference data (the "ground truth" for the software's purpose) was established by the NHANES III study methodologies. This would involve:
- Standardized DXA scanning protocols for BMD measurements across various skeletal sites.
- Strict inclusion/exclusion criteria for subjects.
- Demographic categorization (sex, age, ethnicity) of the subjects.
- Statistical analysis of the collected BMD data to derive mean values and standard deviations for each demographic stratum and skeletal region.
- Expert review and publication of these statistical findings in a peer-reviewed journal (A.C. Looker, et al., 1995).
- Again, not a traditional "training set" for an AI model. The reference data (the "ground truth" for the software's purpose) was established by the NHANES III study methodologies. This would involve:
§ 892.1170 Bone densitometer.
(a)
Identification. A bone densitometer is a device intended for medical purposes to measure bone density and mineral content by x-ray or gamma ray transmission measurements through the bone and adjacent tissues. This generic type of device may include signal analysis and display equipment, patient and equipment supports, component parts, and accessories.(b)
Classification. Class II.