OPTIMIZED EFFECTIVENESS BASED SLEEP AID MANAGEMENT

Abstract

A melatonin optimization system detects users' hormone sensitivity through sleep architecture monitoring and recommends a dose of melatonin personalized to each user in relation to the user's behaviors, needs and health conditions. Determining an optimized melatonin dose requires accurate prediction of non-intervention sleep onset latency for the upcoming sleep period so that the dose can be based on the difference between the user's desired sleep onset latency and the predicted non-intervention sleep onset latency. The system can use either a general population-based sleep onset latency prediction model or a machine learning model trained to be personalized for each user.

Claims

1. A melatonin optimization system for optimizing the effectiveness of exogenous melatonin in achieving a desired sleep outcome for a user, the system comprising: a user interface configured to accept information input to the user interface regarding health conditions, self-reported behavior, and a desired sleep outcome of the user; a sleep architecture detection module configured to perform monitoring of a sleep architecture of the user and to detect a hormone sensitivity of the user through the monitoring; a behavior detection module configured to detect and collect information about behavior of the user in order to define a detected behavior of the user; an initial dose algorithm module configured to define an initial advised dose of melatonin for the user; an effectiveness evaluation module configured to determine an outcome difference between the desired sleep outcome of the user and a measured sleep outcome of the user; and a recommendation engine configured to define an intervention for the user to reduce the outcome difference, wherein the initial dose algorithm module is configured to define the initial advised dose of melatonin based on the information input to the user interface, and wherein the recommendation engine is configured to define the intervention based on the outcome difference, the monitoring of the sleep architecture, the detected behavior of the user, and the initial advised dose of melatonin.

2. The melatonin optimization system of claim 1, wherein the desired sleep outcome is a sleep onset latency.

3. The melatonin optimization system of claim 1, wherein the self-reported behavior includes information about food intake, alcohol intake, and caffeine intake.

4. The melatonin optimization system of claim 1, wherein the behavior detection module includes a stress detector configured to detect user physiological data comprising at least one of a heart rate variability and a skin conductance.

5. The melatonin optimization system of claim 1, wherein the behavior detection module includes a stress detector, the stress detector comprising a device configured to communicate via an application programming interface (API) with remote stress detection software, wherein the remote stress detection software is configured to determine a stress level of the user.

6. The melatonin optimization system of claim 1, wherein the behavior detection module includes a light sensor configured to determine the duration, intensity, and timing of both sunlight and artificial light to which a user is exposed throughout the day and up to the user's bed time.

7. The melatonin optimization system of claim 1, wherein the intervention includes a change to a current dose of melatonin being recommended to the user, wherein the change to the current dose of melatonin being recommended to the user is based on: an effectiveness of any previous melatonin intervention, the user's behavior for the current day as detected by the behavior detection module, and the user's sleep architecture from the previous night as detected by the sleep architecture detection module.

8. The melatonin optimization system of claim 1, wherein, if a current recommended melatonin dose has reached a predetermined maximum level, the recommendation engine will define the intervention to only include changes to the user's behavior.

9. The melatonin optimization system of claim 1, wherein the effectiveness evaluation module comprises a machine learning model, wherein the machine learning model is trained to provide a non-intervention sleep onset latency prediction for an upcoming sleep period of the user based on data collected by the sleep architecture detection module regarding a most recent sleep period of the user.

10. The melatonin optimization system of claim 9, wherein the machine learning model has been provided training to personalize the non-intervention sleep onset latency prediction for the user, wherein the training of the machine learning model comprises providing sleep architecture data of the user from a baseline period when the user was not using exogenous melatonin to the machine learning model and providing sleep architecture data of the user from a testing period when the user was using varying doses of exogenous melatonin to the machine learning model.

11. The melatonin optimization system of claim 9, wherein the training of the machine learning model further comprises providing behavior data of the user from the behavior detection module associated with the baseline period to the machine learning model and providing behavior data of the user from the behavior detection module associated with the testing period to the machine learning model.

12. The melatonin optimization system of claim 9, wherein the machine learning model has been provided training to predict a non-intervention sleep onset latency for the user, wherein the training of the machine learning model comprises providing sleep architecture data of research study subjects from a baseline period when the research study subjects were not using exogenous melatonin to the machine learning model and providing sleep architecture data of the research study subjects from a testing period when the research study subjects were using varying doses of exogenous melatonin to the machine learning model.

13. The melatonin optimization system of claim 9, wherein the recommendation engine is configured to define the intervention based on the non-intervention sleep onset latency prediction provided by the machine learning model.

14. The melatonin optimization system of claim 10, wherein the recommendation engine is configured to define the intervention based on the non-intervention sleep onset latency prediction provided by the machine learning model.

15. The melatonin optimization system of claim 12, wherein the recommendation engine is configured to define the intervention based on the non-intervention sleep onset latency prediction provided by the machine learning model.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0008] FIG. 1 shows a schematic depiction of a melatonin optimization system, according to an exemplary embodiment of the disclosed concept;

[0009] FIG. 2 shows a schematic depiction of a more detailed variation of the melatonin optimization system depicted in FIG. 1, according to an exemplary embodiment of the disclosed concept;

[0010] FIG. 3 shows a graph illustrating a two-process model of sleep alertness;

[0011] FIG. 4 is a schematic depiction of a multi-modal input sleep onset latency prediction module that can be included in an effectiveness evaluation module of either of the systems depicted in FIG. 1 and FIG. 2, according to an exemplary embodiment of the disclosed concept; and

[0012] FIG. 5 is a flow chart containing the steps of a method for training a machine learning model to predict sleep onset latency for a user of either of the systems depicted in FIG. 1 and FIG. 2, according to an exemplary embodiment of the disclosed concept.

DETAILED DESCRIPTION OF EXEMPLARY EMBODIMENTS

[0013] As used herein, the singular form of “a”, “an”, and “the” include plural references unless the context clearly dictates otherwise.

[0014] As used herein, the statement that two or more parts or components are “coupled” shall mean that the parts are joined or operate together either directly or indirectly, i.e., through one or more intermediate parts or components, so long as a link occurs.

[0015] As used herein, the term “number” shall mean one or an integer greater than one (i.e., a plurality).

[0016] As used herein, the term “controller” shall mean a number of programmable analog and/or digital devices (including an associated memory part or portion) that can store, retrieve, execute and process data (e.g., software routines and/or information used by such routines), including, without limitation, a field programmable gate array (FPGA), a complex programmable logic device (CPLD), a programmable system on a chip (PSOC), an application specific integrated circuit (ASIC), a microprocessor, a microcontroller, a programmable logic controller, or any other suitable processing device or apparatus. The memory portion can be any one or more of a variety of types of internal and/or external storage media such as, without limitation, RAM, ROM, EPROM(s), EEPROM(s), FLASH, and the like that provide a storage register, i.e., a non-transitory machine readable medium, for data and program code storage such as in the fashion of an internal storage area of a computer, and can be volatile memory or nonvolatile memory.

[0017] As used herein, the term “intervention” shall refer to a dosage of exogenous melatonin and/or a set of behaviors recommended for a person seeking to change a number of characteristics of his or her sleep.

[0018] As used herein, the term “machine learning model” shall mean a software system that develops and builds a mathematical model based on sample data, known as “training data”, in order to make predictions or decisions without being explicitly programmed to do so, including, without limitation, a computer software system that develops that has been trained to recognize patterns from a set of training data, and subsequently develops algorithms to recognize patterns from the training data set in other data sets.

[0019] Directional phrases used herein, such as, for example and without limitation, top, bottom, left, right, upper, lower, front, back, and derivatives thereof, relate to the orientation of the elements shown in the drawings and are not limiting upon the claims unless expressly recited therein.

[0020] The disclosed concept, as described in greater detail herein in connection with various particular exemplary embodiments, provides methods and systems for effectively administering exogenous melatonin so as to meaningfully accelerate the onset of sleep. It should be noted that exogenous melatonin is administered to effect a timing function of sleep rather than a hypnotic effect; that is, exogenous melatonin is administered to influence when a person falls asleep but may not affect the total amount of time that a person sleeps. Research literature shows that responses to melatonin administered exogenously (i.e. ingested) are greatest when at times when endogenous levels (i.e. natural bodily production levels) are not normally present, that is, during the day. Conversely, the effect of taking melatonin during the time when it is already being produced by the body (i.e. during the night) is minimal. When taken in the late night/morning, melatonin causes phase delays (shifts to a later time) in the body's biological night as defined by the body's circadian rhythm, and when ingested in the afternoon or evening, exogenous melatonin causes phase advances (shifts to an earlier time) in the body's biological night as defined by the body's circadian rhythm.

[0021] The conditions during which melatonin is administered appear to be very important and may dictate the effectiveness of any given dose, particularly with respect to acute changes in core body temperature (CBT) and sleepiness. Accordingly, the disclosed concept provides a system for optimizing the effectiveness of melatonin treatment by detecting hormone sensitivity through sleep architecture monitoring and personalizing dosing in relation to users' behaviors, needs and health conditions. As used herein, the terms “dose”, “dosage”, and “dosing” encompass both the quantity of melatonin to be taken and the timing of ingestion of the melatonin relative to a desired sleep event or time of day.

[0022] FIG. 1 is a schematic depiction of a melatonin optimization system 10 for maximizing melatonin effectiveness based on a user's needs, characteristics, and behavior, in accordance with an exemplary embodiment of the disclosed concept, and FIG. 2 is a schematic depiction of a system 100 that is a more detailed variation of the system 10. FIG. 2 also depicts some of the decision making executed by the components of systems 10 and 100. It should be noted that a behavior detection module 15 is depicted as being separate from a recommendation engine 16 in FIG. 1 while being depicted as part of the recommendation engine 16 in FIG. 2, and that either configuration can be used without departing from the scope of the disclosed concept. The system 10, 100 collects information about a user's behavior, characteristics, sleep architecture, and desired sleep outcomes as inputs to the system, then analyzes the inputs to make recommendations to help the user achieve the desired sleep outcomes. In an exemplary embodiment, the data collection and analysis tasks of the systems 10, 100 can be executed by any type of controller or computing system with input/output, processing, and memory capabilities, and any number or combination of such controller or computing systems can be used without departing from the scope of the disclosed concept.

[0023] Referring to FIGS. 1 and 2, a user interface 11 allows the user to provide information about his or her age, gender, usual bed time, timing and dosage of caffeine and alcohol intake, health conditions (e.g. high blood pressure, insomnia, diabetes, beta-blocker intake, jet-lag), and desired sleep outcome(s) (i.e. sleep onset latency, sleep/wake time, sleep duration, etc.) to the system 10, 100. Such user interface 11 can, for example and without limitation, take the form of a mobile phone application or internet-based portal. An initial dose algorithm module 12 provides initial melatonin dose and timing recommendations for the user based on the age/gender/conditions/needs/desired sleep time data entered by the user into the user interface 11. The initial dose algorithm module 12 can, for example and without limitation, be preprogrammed to utilize widely available data about recommended dosing based on problems recognized in the general population. Table 1 below provides a non-limiting example of widely available data providing a number of melatonin dosages that are recommended based on a user's specific conditions, needs, desired sleep outcomes, etc.:

TABLE-US-00001 TABLE 1 Dosage of Melatonin Recommended for Specified Conditions User Condition Recommended Melatonin Dosage(s) for an Adult User has any one of a number of disorders 0.5 mg to 5 mg of melatonin taken daily before bedtime for up that affects when a person sleeps and when to 6 years, as has been used for blind people he/she is awake High dose of 10 mg melatonin taken 1 hour before bedtime for up to 9 weeks, as has been used for blind people 2 mg to 12 mg of melatonin taken at bedtime for up to four weeks User has trouble falling asleep at a 0.3 mg to 5 mg of melatonin daily for up to 9 months conventional bedtime (delayed sleep phase syndrome) User has sleep disturbance caused by 2.5 mg of melatonin daily for up to 4 weeks certain blood pressure medicine (beta Single doses of 5 mg blocker-induced insomnia) User has endometriosis 10 mg of melatonin daily for up to 8 weeks User has high blood pressure 2 mg to 3 mg of controlled-release melatonin daily for to 4 weeks User has insomnia 2 mg to 3 mg of melatonin daily before bedtime for up to 29 weeks Higher doses of up to 12 mg daily for shorter durations (up to 4 weeks)

[0024] A sleep architecture detection module 13 includes a number of sensors and/or trackers and a number of algorithms for detecting the user's sleep architecture. Any type of sensor, tracker, or other device/method for collecting sleep architecture can be used without departing from the disclosed concept. Non-limiting examples of devices that can be used to detect sleep architecture include wrist worn devices, mattresses with sleep trackers, and user-entered sleep diaries. An effectiveness evaluation module 14 compares the user's desired sleep outcome (as indicated by the user's input to user interface 11) to the actual sleep outcome (as detected by the sleep architecture detection module 13), and a recommendation engine 16 (described in more detail below) then directs the user to either change the dosage/timing of the melatonin intake or execute a different intervention for changing a behavioral aspect based on the findings of the effectiveness evaluation module 14.

[0025] A behavior detection module 15 collects and processes sensor data and the user's self-reported information (input to the user interface 11) to detect information about the user's behavior, including but not limited to the user's exposure to light, physical activity, food intake, and stress level. In an exemplary embodiment of the disclosed concept, the behavior detection module 15 includes, at a minimum: a wearable sensor for light exposure detection, a stress detector, an activity tracker, and an input from the user interface 11 for self-reporting of food/alcohol/caffeine intake. The wearable sensor for light exposure detection can, for example and without limitation, detect and collect information about the duration, intensity and timing of both sun and artificial light to which a user is exposed throughout the day and up to the user's bed time. The stress detector can, for example and without limitation, be a wearable device that detects user physiological data such as heart rate variability and/or skin conductance. The assessment of stress level can be performed either by an on-device algorithm or by a third party via an application programming interface (API) call from the stress detector. The activity tracker can, for example and without limitation, be a wearable device that detects the number of minutes that the user is engaged in an aerobic activity, with aerobic activity being characterized by the user's heart rate level reaching between 55% and 85% of the user's maximum heart rate, the maximum heart rate (maxHR) being calculated using Equation (1) below:

maxHR=207−0.7*(age of user)  (1)

The user interface for self-reporting of food/alcohol/caffeine intake enables a user to report the amount and timing of his or her food, alcohol, and/or caffeine intake throughout the day.

[0026] The recommendation engine 16 comprises a melatonin dose adjustment module 17 and a behavioral changes module 18 that evaluates the melatonin dose effectiveness and analyzes the detected user behavior along with the user's stated preferences to recommend melatonin dosing adjustments and/or behavioral changes for the user. More specifically, the dose adjustment module 17 determines the difference between the user's desired sleep onset latency and measured sleep onset latency (or any other chosen sleep metric), and recommends an intervention based on the effectiveness of any previous melatonin intervention, the current day's activities, and the prior night's sleep. If the recommended intervention is melatonin dosing and the previous melatonin dose already reached a predetermined maximum allowed level, then only behavioral changes will be recommended. The behavioral changes module 18 computes the difference between recommended behavior for the user and the user's measured behavior. If the behavior difference exceeds a pre-defined threshold, then an intervention is recommended.

[0027] Behavioral change recommendations are provided only for the actual behaviors that are being monitored. For example and without limitation, if the particular implementation of a system 10 being used does not include a light sensor, then the behavioral changes module 18 will not provide light exposure recommendations. A non-limiting list of behavioral change recommendations that can be provided by the behavioral changes module 18 includes: directing the user to engage in an outdoor activity such as walking for at least 30 minutes a day, directing the user to turn off artificial lights at least two hours before the desired time of sleep onset, directing the user to engage in breathing exercises to assist in reducing the user's stress level, and directing the user to avoid caffeine intake after a specified time of day.

[0028] Referring once more to the effectiveness module 14 of systems 10, 100, in addition to comparing the user's desired sleep outcome to the actual sleep outcome, the effectiveness module 14 can additionally predict non-intervention sleep onset (i.e. naturally occurring sleep onset) using a number of statistical or empirical techniques, and use the predicted non-intervention sleep onset as part of the evaluation of the effectiveness of any current intervention. The predicted non-intervention sleep onset can then be used by the recommendation engine 16 to determine a recommended intervention for the upcoming night (or other period of sleep). Referring to FIG. 3, one non-limiting example of a technique that the effectiveness module 14 can employ for predicting sleep onset is a two-process model of sleep alertness. The two-process model of sleep alertness uses time since awakening, time since falling asleep, and time of day to generate a sleep-dependent curve (Process S) and sleep-independent circadian curve (Process C). The Process S curve demonstrates that alertness is highest upon waking and decreases steadily throughout the day, reaching its lowest level at sleep onset (represented by the lowest value on the Process S curve). The Process S reverses at sleep onset and is called S′ until awakening. The S′ portion of the Process S curve shows that homeostatic sleep debt is relieved during sleep.

[0029] Still referring to FIG. 3, a predicted alertness level for the time of day is expressed as the arithmetic sum of the S and C curves (depicted as a dashed line S+C in FIG. 3). Sleep onset latency (SOL) for any given time of day can then be predicted using Equation (2) below:

y=0.56*10.sup.0.12x  (2)

where x is the predicted alertness for a specified time of day (i.e. found by locating the S+C curve point corresponding to the time of day) and y is the sleep onset latency. It will be appreciated that the constants 0.56 and 0.12 in Equation (2) result from using known parameter estimation methods, wherein these constants were found by defining a best-fit relationship between observed alertness level and sleep onset latency of several research study test subjects. More specifically, for a lowest level of predicted alertness where x=1, it was found that the sleep latency y 0.5 minutes.

[0030] The effectiveness module 14 can include a SOL prediction module 140 incorporating a machine learning model 145 (the SOL prediction module 140 and machine learning model 145 both being described in more detail with respect to FIGS. 4 and 5) to predict sleep onset latency n hours before the user's desired bedtime. The user's preferred bedtime can either be expressed by the user through the user interface 11 or can be automatically detected from historical sleep architecture data collected by the sleep architecture detection module 13, by averaging bedtime detected over a week of use, and averaging bedtime separately over weekdays and weekends if different sleep routines are present in those two parts of the week. In a non-limiting exemplary embodiment of the disclosed concept, rather than predicting sleep onset latency using Equation (2) above, which is deemed applicable to the general population based on data compiled from several research subjects, the machine learning model 145 can be configured to implement a personalized SOL prediction tailored to a particular user of the system 10, 100.

[0031] Referring to FIG. 4, which shows a multi-modal input SOL prediction module 140 that includes the machine learning model 145, and FIG. 5, which shows the steps of a method 200 for training the machine learning model 145 to predict sleep onset latency, the machine learning model 145 can be trained to make user-personalized SOL predictions by analyzing sleep architecture data collected by the sleep architecture detection module 13. The effect of training the machine learning model 145 to make personalized SOL predictions can be to find constants more personalized to the user than the constants 0.56 and 0.12 used in Equation (2), i.e. y=0.56*10.sup.0.12x, and even to define a relationship other than the one defined in Equation (2) to predict sleep onset latency for the user. The accuracy of a personalized SOL prediction function that allows for multiple inputs (as the multi-modal input SOL prediction module 140 does) can be maximized by minimizing the value of an error function defined in Equation (3) below:

error(X)=Σ.sub.i=0.sup.K(measuredSOL(i)−predictedSOL(i)(X)).sup.2  (3)

wherein X represents all needed parameters (all such parameters being the parameters chosen as input to the multi-modal input SOL prediction module 140), and K is the number of days in the training phase. It will be appreciated that a number of known techniques for solving the minimization of an error function exist, for example and without limitation the gradient descent technique, and that any known optimization algorithm or other technique for minimizing the error function defined in Equation (3) can be used without departing from the scope of the disclosed concept.

[0032] Still referring to FIGS. 4 and 5, to train the machine learning model 145 to predict sleep onset latency for a user, a new user is first directed at step 201 of the method 200 to use the system 10 or system 100 for a baseline period (e.g. one week) without taking any melatonin. Next, the user is directed at step 202 to take varying doses of melatonin throughout the course of a subsequent testing period (e.g. one week), wherein the doses are defined as either minimum, maximum, or intermediate (e.g. the intermediate dose being defined as halfway between the maximum dose and minimum dose). Together, the baseline period and the testing period comprise a data collection period. It will be appreciated that the durations of the baseline period and the testing period can be longer or shorter than one week, and that the durations of the baseline period and the testing period can vary from one another without departing from the scope of the disclosed concept. The testing period or baseline period for the user can also be omitted entirely without departing from the disclosed concept such that the machine learning model 145 can be trained using comparable available data from dedicated research trials wherein sleep architecture data was collected for multiple research subjects who underwent a baseline period and a testing period; however, it will be appreciated that the SOL prediction made by the machine learning model 145 in that instance would not be personalized for the user of the system 10 or 100.

[0033] In FIG. 4, the SOL prediction module 140 is depicted as including a two-process (2-P) SOL prediction module. The 2-P SOL prediction module uses an algorithm that takes the two processes of homeostatic sleep drive and circadian timing (i.e. the Process S curve and the Process C curve, respectively, shown in FIG. 3) into account in predicting sleep onset latency. However, it will be appreciated that the Unified Model of Performance, the SAFTE model, or any other biological model that takes into account homeostatic sleep drive and circadian timing can be used in place of the 2-P model in the SOL prediction module 140 without departing from the scope of the disclosed concept. Referring to FIG. 5, a determination is made at step 203 about whether a multi-modal SOL prediction will be sought in step 205, i.e. a SOL prediction that takes behavior into account in addition to homeostatic sleep drive and circadian timing. If a multi-modal SOL will be sought, then data collected by the behavior detection module 15 about the user's behavior is provided to the machine learning model 145 at step 204. If a multi-modal SOL prediction is not sought, the process 200 advances to step 205 from step 203.

[0034] At step 205, the machine learning model 145 analyzes the baseline period data, the testing period data, and the behavior detection module 15 data (if applicable) for each day of the data collection period and compares the data to the user's detected sleep architecture collected every day of the data collection period by the sleep architecture detection module 13 in order to determine patterns indicative of SOL. As previously stated with respect to steps 201 and 202 of process 200, if a personalized SOL prediction is not sought, user baseline data and testing data may not be provided to the machine learning model 145 at step 205, and research subject baseline data and testing data are instead provided so that the machine learning model 145 can compare the research subjects' baseline data and testing data to the research subjects' sleep architecture data in order to determine patterns indicative of SOL.

[0035] Once the machine learning model 145 has been trained using method 200 to recognize patterns and associations between sleep architecture data and baseline period data, testing period data, and behavior data (if applicable), the machine learning model 145 is capable of making a reliable SOL prediction based on newly provided data about the user's sleep architecture from the previous night (or other sleeping period), the most recent melatonin dose taken by the user, and the user's most recent behavior (if applicable). It will be appreciated that several techniques are available for building a machine learning model such as the machine learning model 145 used in the SOL prediction module 140, including but not limited to decision tree regressors, generalized linear models, and other regression modeling task solutions, and that any technique for building a machine learning model may be used to build machine learning model 145 without departing from the scope of the disclosed concept.

[0036] In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word “comprising” or “including” does not exclude the presence of elements or steps other than those listed in a claim. In a device claim enumerating several means, several of these means may be embodied by one and the same item of hardware. The word “a” or “an” preceding an element does not exclude the presence of a plurality of such elements. In any device claim enumerating several means, several of these means may be embodied by one and the same item of hardware. The mere fact that certain elements are recited in mutually different dependent claims does not indicate that these elements cannot be used in combination.

[0037] Although the invention has been described in detail for the purpose of illustration based on what is currently considered to be the most practical and preferred embodiments, it is to be understood that such detail is solely for that purpose and that the invention is not limited to the disclosed embodiments, but, on the contrary, is intended to cover modifications and equivalent arrangements that are within the spirit and scope of the appended claims. For example, it is to be understood that the present invention contemplates that, to the extent possible, one or more features of any embodiment can be combined with one or more features of any other embodiment.