System and method for developing a research protocol
Patent Information
- Application Number
- HK32026122583
- Authority / Receiving Office
- HK · HK
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2026-04-27
- Publication Date
- 2026-08-28
Abstract
Description
SYSTEM AND METHOD FOR DEVELOPING A RESEARCH PROTOCOL 5 10 15 20 25 30 FIELD OF THE DISCLOSURE The present disclosure relates to a computing method and system for developing / devising a research protocol, preferably using an artificial intelligence based model configured to provide guidance in the development of said research protocol. BACKGROUND OF THE DISCLOSURE Subject matter experts in research disciplines typically have deep technical knowledge of their specialised field but often lack the same level of knowledge in adjacent and essential fields such as statistical analyses, research protocol design and the like. Formulating and conceptualising a clinical research study requires significant time commitment by the researcher; and often one or more consultations with a statistical expert to ensure that the research study is statistically robust prior to actually undertaking any study. With the rapid evolution in underlying technical capacity and sophistication of generative AI models and natural language processing provides increasingly sophisticated applications some researchers have attempted to work with AI models to assist in their research study design. However, notwithstanding the rapid development of underlying models and natural language processing, there are numerous issues or problems which limit their usefulness in this context. The output from generative artificial intelligence models is generally based on mimicry and statistical inference derived from the training data upon which the artificial intelligence model is trained. Many models lack appropriate context especially in specialised domains such as clinical research. Models may provide inaccurate results, results which lack robustness and may be unreliable, confidently “hallucinating” and providing entirely inaccurate answers especially when responding to complex queries. Typically, clinical research subject matter experts have limited experience in prompt engineering and interaction with artificial intelligence models; and may be reluctant to engage via chatbot or the like with such generative AI models if required to provide overly complex or predetermined prompting materials. Furthermore, time poor professionals such researchers are unlikely to persist in interacting with the model if the interaction does not immediately appear useful and tailored to their specific requirements. Accordingly, subject matter experts from highly specialised domains such as medical research remain highly sceptical of the usefulness of such tools; and are unlikely to engage after an initial high level (unsuccessful) interaction. 1 HK 30136966 A 5 10 15 20 25 30 It is an object of the present disclosure to provide a method and system of developing a research protocol using a computing platform executing an artificial intelligence model which addresses or at least ameliorates some of the above issues; or at least provides the public with a further choice. SUMMARY OF THE DISCLOSURE Features and advantages of the disclosure will be set forth in the description which follows, and in part will be obvious from the description, or can be learned by practice of the herein disclosed principles. The features and advantages of the disclosure can be realized and obtained by means of the instruments and combinations particularly pointed out in the appended claims. In accordance with a first aspect of the present disclosure, there is provided a method of developing a research protocol by a processor of a computing platform executing an artificial intelligence model; the method comprising the sequentially performed steps of: receiving by the processor a first user input from a client side device operable by the user comprising a research idea; generating a first output by the processor from the research idea; wherein the first output comprises a proposed question for a research study; receiving confirmation of the first output from the user; generating a second output by the processor from the first output wherein the second output comprises a research study framework and outcome addressing the proposed question, said research study framework having a predetermined type; receiving by the processor confirmation of the second output from the user; generating by the processor a third output from the research study framework and outcome, wherein said third output comprises a recommendation of the statistical parameters for the research protocol. Optionally, the method may further comprise verifying by the model executing on the processor the third output against standard checklists corresponding to said type of research study, or in the absence of a standard checklist identifying further information suggested. Preferably, the standard checklists may be selected according to the type of research study from a group comprising checklists for observational studies, systematic review and meta- analysis, interventional studies and diagnostic studies. Preferably, the first output is generated by the processor using a machine learning model configured such that if the research idea is determined to be about comparing intervention or 2 HK 30136966 A 3 exposures the research question is formulated using a Population Intervention Comparison and Outcome (PICO) framework. If the research question is formulated as a diagnostic test an intermediate output may be generated by the artificial intelligence model to obtain user input on the outcome of interest, user input on a standard benchmark test for measuring the outcome of the interest, and user 5 input parameters required for execution of said diagnostic test. If the research premise is formulated as an epidemiological study; observational study or study involving intervention, an intermediate output may be generated for confirmation by the user of the outcome under evaluation, exposure or risk factor to be evaluated. If the research premise is formulated as an epidemiological study type; observational study 10 type or interventional study type, the second output further comprises a suggested suitable data type for confirmation. Advantageously, the third output comprises one or more of guidance on the procedures for testing the difference in baseline variables, statistical assumptions required, steps to test statistical assumptions and sample size calculation. 15 Preferably, each of outputs are only generated once the preceding step has been completed. Preferably, the outputs for each step are generated from stored earlier decisions and assumptions using an artificial intelligence model which is a Large Language Model (LLM). Advantageously, the artificial intelligence model which is a Large Language Model (LLM) which is configured by an external control architecture which requires that each of outputs are 20 only generated once the preceding step has been completed and stores state information for each step. Preferably, the method may further comprise generating a research protocol in accordance with a predetermined framework. In a further aspect, there is provided a computer readable code product which when executed 25 by a processor causes an artificial intelligence model executing on the processor to perform the steps described herein. In a further aspect, there is described a system for developing a research protocol comprising a processor of a computing platform executing an artificial intelligence model comprising: a proposed question generation module configured for generating a first output comprising a 30 research question from a research idea received from the user; HK 30136966 A 5 10 15 20 25 30 a research study framework generation module for generating a second output comprising a research study framework and outcome from the first output, said research study framework having a predetermined type; a statistical parameter recommendation module for generating a third output from the second output; wherein said third output comprises a recommendation of the statistical parameters for the research protocol. Optionally, the system further comprises a verification module for verifying the third output against standard checklists corresponding to said type of research study, or in the absence of a standard checklist identifying further information suggested. Preferably, the system comprises an external control architecture which ensures that outputs of each module are generated sequentially only upon execution by the preceding module and user confirmation thereof, and wherein said framework is configured for ensuring state information for each module is stored. Advantageously, each of the modules are executed on an artificial intelligence model which is a Large Language Model (LLM) which is configured by an external control architecture which requires that each of modules operate sequentially once a preceding module has generated an output and wherein state information for each module is stored. BRIEF DESCRIPTION OF THE DRAWINGS In order to describe the manner in which the above-recited and other advantages and features of the disclosure can be obtained, a more particular description of the principles briefly described above will be rendered by reference to specific embodiments thereof which are illustrated in the appended Figures. Understanding that these Figures depict only exemplary embodiments of the disclosure and are not therefore to be considered to be limiting of its scope, the principles herein are described and explained with additional specificity and detail through the use of the accompanying Figures. Preferred embodiments of the present disclosure will be explained in further detail below by way of examples and with reference to the accompanying Figures, in which:- Fig 1A depicts an exemplary schematic view of an embodiment of the system of the present disclosure. Fig 1B depicts an exemplary flow diagram showing steps according to an embodiment of the present disclosure. Fig 1C depicts an alternate exemplary flow diagram showing additional steps as compared with the flow diagram depicted in Fig 1B. 4 HK 30136966 A 5 10 15 20 25 30 DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Various embodiments of the disclosure are discussed in detail below. While specific implementations are discussed, it should be understood that this is done for illustration purposes only. A person skilled in the relevant art will recognize that other components and configurations may be used without departing from the spirit and scope of the disclosure. The disclosed technology addresses the need in the art for assisting subject matter experts in clinical and research disciplines to develop robust research study protocols. Referring to the Figures, there is depicted in Fig 1A an exemplary schematic view of an embodiment of the system 10 of the present disclosure. As depicted, there is shown a server 20 including one or more processors 22 which execute an artificial intelligence model 24. Preferably, the artificial intelligence model 24 is a trained machine learning model, and more preferably, a generative artificial intelligence model 24 such as large language model (LLM). As is known in the art, an LLM model may contain millions or billions of learned parameters which enable it to model how words relate to each other in a textual sequence, based on probabilistic assessment of relationships between words. Exemplary models that the present disclosure may be used with LLM models which include Phi-4, Deepseek and Gemini 2.5 Pro without limitation. As is known in the art, these models are language models that have been trained (in an unsupervised manner) on a large corpus derived from documents available to the public online. These models have a very large number of learned parameters and hence are able to accept a large number of tokens as input and able to generate a large number of tokens as output. The present disclosure is especially suitable for use with a generative pre-trained transformer LLM model which are specialised types of LLM capable of generating human-like text in a sequential generation process. An exemplary generative pre-trained transformer LLM model, is GPT model 3, GPT- model 4 and GPT-model 5. GPT-models trained as a generative models can process input text sequences to predictively generate a meaningful output text sequence. ChatGPT is built on top of a GPT-type LLM, and has been fine-tuned with training datasets based on text-based chats (e.g., chatbot conversations). ChatGPT is designed for processing natural language, receiving chat-like inputs and generating chat-like outputs Advantageously, the parameters underlying the performance of the AI models may be stored on a data store 26 which is hosted on (or in communication with) server 20. Preferably, the server 20 is accessible via a communications interface 28 across a communications network 30 to a user operable device 32 which may be remotely located from the server 20. 5 HK 30136966 A 6 Notably, the artificial intelligence model 24 may execute on one or a plurality of processors 22 (e.g., hardware processors such as, for example, processors of cooperating computer systems). Indeed, processing of inputs by an LLM may be computationally expensive / may involve a large number of operations (e.g., many instructions may be executed / large data structures may be accessed from memory) and providing output in a required timeframe (e.g., 5 real-time or near real-time) may require the use of a plurality of processors / cooperating computing devices as discussed above. Exemplary sub systems may be arranged on the server 20 as depicted in the form of modules as shown. Specifically, a research question module 25a, research study framework module 25b, statistical module 25c. Similar or additional modules may be included without limitation. 10 In particular, preferably, the system 10 for developing a research protocol comprising a processor of a computing platform executing an artificial intelligence model 24 comprises a research question module 25a configured for generating a first output comprising a research question from a research idea received from the user; a research study framework module 25b module for generating a second output comprising a 15 research study framework and outcome from the first output, said research study framework having a predetermined type; a statistical module 25c (parameter recommendation) for generating a third output from the second output; wherein said third output comprises a recommendation of the statistical parameters for the research protocol. 20 Optionally, the system 10 may further comprise a verification module 25d for verifying the third output against standard checklists corresponding to said type of research study, or in the absence of a standard checklist identifying further information suggested. Advantageously, the system 10 may be configured such that it includes an external control architecture 29 which ensures that outputs of each module are generated sequentially only 25 upon execution by the preceding module, and wherein said framework is configured for ensuring state information for each module is stored. Preferably, each of the modules are executed on an artificial intelligence model 24 which is a Large Language Model (LLM) which is configured by an external control architecture 29 which requires that each of modules operate sequentially once a preceding module has generated 30 an output and wherein state information for each module is stored. The user device 32 may be configured to receive natural language inputs from a user, processor them on processor 33; and transmit these inputs to the server 20 across the network HK 30136966 A 7 so as to cause it to generate via the AI model 24 an output. The natural language inputs may be use in the generation of prompts. A computing system may generate a prompt that is provided as input to the LLM via its API. The prompt may optionally be processed or pre-processed into a token sequence prior to being provided as input to the LLM via its API. A prompt can include one or more examples 5 of the desired output, which provides the LLM with additional information to enable the LLM to better generate output according to the desired output. Additionally, or alternatively, the examples included in a prompt may provide inputs (e.g., example inputs) corresponding to / as may be expected to result in the desired outputs provided. A one-shot prompt refers to a prompt that includes one example, and a few-shot prompt refers 10 to a prompt that includes multiple examples. Preferably, the AI model is configured to receive user input via an interface and Referring now to Figure 1B, there is provided an exemplary flow diagram which shows steps according to an embodiment of the method of developing a research protocol by a processor of a computing platform executing an artificial intelligence model of the present disclosure. 15 As depicted, the method comprises the sequentially performed steps of: receiving by the processor a first user input from a user operable device 32 by the user comprising a research idea in Step S1; generating a first output by the processor from the research idea; wherein the first output comprises a proposed question for a research study in Step S2; 20 receiving confirmation of the first output the user in Step S3; generating a second output by the processor from the first output wherein the second output comprises a research study framework and outcome addressing the proposed question, said research study framework having a predetermined type in Step S4; receiving by the processor confirmation of the second output from the user in Step S5; 25 generating by the processor a third output from the research study framework and outcome, wherein said third output comprises a recommendation of the statistical parameters for the research protocol in S6. Preferably, the confirmation of the output in Step S3 and Step S5 is provided by the user via the user operable device, either directly or via an API to the server across the communication 30 network. HK 30136966 A 5 10 15 20 25 30 Optionally, in another aspect there is provided computer readable code product which when executed by a processor causes an artificial intelligence model executing on the processor to perform the steps of the method set out above. In an example which is referred to consistently hereinafter, the user input provided in Step S1 may be a research idea of studying whether smoking causes cancer. Preferably, in Step S1, the first output is generated by the processor using a machine learning model configured such that if the research idea is determined to be about comparing intervention or exposures the research question is formulated using a Population Intervention Comparison and Outcome (PICO) framework. If not applicable, categorize their question into whether it belongs to epidemiological studies, qualitative study as described below. In the case of comparing intervention or exposures the research question is formulated using a Population Intervention Comparison and Outcome (PICO) framework is applicable. The PICO framework stands for: 1. P - Population: Who are we studying? 2. I - Intervention or Exposure: What is the factor we are interested in? (In this case, smoking) 3. C - Comparison: What is the alternative to the exposure? (In this case, not smoking) 4. O - Outcome: What is the health outcome we are measuring? (In this case, cancer) In an exemplary embodiment, using this framework the research question based on the user input of a proposed study into cancer may be formulated as: 1. P (Population): Who are you interested in studying? The general population? A specific age group (e.g., adults over 40)? People in a certain geographic area? A clear definition of the study population is crucial. 2. I (Exposure): How will we define "smoking"? This needs to be precise. For example, are we interested in: o Current smokers vs. former smokers vs. never-smokers? o The quantity of smoking? We often measure this in "pack-years" (number of packs smoked per day multiplied by the number of years the person has smoked). o The type of smoking (e.g., cigarettes, cigars, vaping)? 3. C (Comparison): The natural comparison group would be "non-smokers." It is 8 HK 30136966 A 5 10 15 20 25 30 often best to be specific and define this group as "never-smokers" to avoid ambiguity with former smokers. 4. O (Outcome): "Cancer" is a very broad category. Smoking is a known risk factor for many types of cancer (e.g., lung, bladder, oesophageal, pancreatic, throat). For a single study, it is usually best to focus on one specific type of cancer. The most strongly associated is lung cancer. Would you like to focus on lung cancer, or is there another type you are interested in? This framework may be used to generate an exemplary research question in Step S2 as a first output by the AI model. Continuing with the Example provided the exemplary research question based on the research idea of smoking provided by the user may be the following: In adults aged 40-75 (P), is current cigarette smoking (I) compared to never smoking (C) associated with a higher incidence of lung cancer (O)?" After generation of the research question, a summary of the question and progress may be provided by the model: Summary of Our Progress • Research Question: To investigate the relationship between smoking and cancer. We are currently refining this using the PICO framework. • Population (P): To be determined. • Intervention / Exposure (I): To be determined (needs specifics on smoking status and quantity). • Comparison (C): To be determined (proposed: never-smokers). • Outcome (O): To be determined (proposed: a specific type of cancer, such as lung cancer). • Study Design: To be determined. • Measures and Data: To be determined. • Statistical Analysis: To be determined. • Sample Size: To be determined. Confirmation from the user of the proposed research question generated by the AI Model executing on the server may be received as depicted in Step S3. Next in Step S4, a second output may be generated by the processor from the first output wherein the second output comprises a research study framework and outcome addressing the proposed question, said 9 HK 30136966 A 5 10 15 20 25 30 research study framework having a predetermined type. Optionally, the advantages and disadvantages of research study framework(s) provided may also be provided. The system may be configured to propose possible study designs and ask user for their choice. If asked, discuss the pros and cons of different possible study designs comprehensively and concisely. Continuing the example: For your question, two designs are most relevant: the Cohort Study and the Case-Control Study. 1. Prospective Cohort Study (Strongest Observational Design for this Question) This study offers temporality, calculates incidence and has less recall bis. However, it is time and cost intensive and requires a large sample size. 2. Case-Control Study (A More Efficient Alternative) This design works in reverse. Start by identifying a group of patients who already have lung cancer (the "cases"). Then, we would select a comparable group of people from the same population who do not have lung cancer (the "controls"). We would then look backwards in time (retrospectively), usually via questionnaires or medical record review, to determine the smoking history of everyone in both groups. We would then compare the odds of having been a smoker in the cases versus the controls. This approach is efficient for rate disease, requires a smaller sample size. However, it is subject to recall bias, selection bias and cannot calculate incidence. Recommendation: For establishing the strongest evidence, a prospective cohort study is the superior design. Landmark studies that established the link between smoking and cancer were indeed large cohort studies (e.g., the British Doctors Study). However, a case-control study is a very practical and valid way to investigate this question, provided it is designed and conducted with extreme care to minimize bias. Which of these designs seems more feasible or appealing for your research plan? < Summary Omitted in Example for brevity and clarity> A summary (not shown) may then be presented for maintaining state information of records completed workflow stages, storing validated decisions and assumptions, tracking unresolved dependencies, and conditions subsequent model outputs based on stored state information. This persistent state enables continuity and consistency across multi-turn interactions and prevents contradictory or redundant outputs. 10 HK 30136966 A 5 10 15 20 25 30 In Step S5, the research study framework and outcome addressing the proposed question may be confirmed by the user. In Step S6, based on the confirmed research study framework and outcome addressing the proposed question, a third output from the research study framework and outcome comprising a recommendation of the statistical parameters for the research protocol may be generated. Such statistical parameters may include the statistical test needed for the primary outcome, the procedures needed to test for differences in baseline variables, statistical assumptions and guide them check the data before embarking on statistical analysis. Stepwise instructions amy also be provided to teach user to check the required underlying assumptions. Finally, the user may be guided on the statistical approach to sample size calculation and suitable formula and appropriate assumptions suggested. Continuing with the example noted above: Calculating the required sample size is a critical step. It ensures your study has enough statistical power to detect a meaningful effect if one truly exists. An underpowered study is a waste of resources and can be unethical, as it exposes participants to potential risks without a high likelihood of answering the research question. Conversely, an overly large study also wastes resources. For a prospective cohort study using a Cox Proportional Hazards model, the sample size calculation is primarily driven by the number of events (i.e., new lung cancer cases) you expect to observe, not just the total number of participants. The Formula A common formula for calculating the required number of events (d) to compare two groups is based on the work of Freedman (1982): d = (Zα / 2 + Zβ)² / (p_smoker * p_never_smoker * (ln(HR))²) … Parameters We Need to Define: 1. Alpha (α) or Significance Level: This is the probability of a Type I error (a false positive). The standard convention in medical research is α = 0.05. This means you accept a 5% chance of concluding there is an association when one doesn't exist. 2. Power (1-β): This is the probability of detecting an association if it truly exists (avoiding a false negative). The standard convention is Power = 0.80 (or 80%). 3. Hazard Ratio (HR): This is the smallest effect size that you consider clinically 11 HK 30136966 A 12 important and want your study to be able to detect. Based on large-scale studies like the Cancer Prevention Study (CPS-II), the hazard ratio for lung cancer in current smokers versus never-smokers is very large, often exceeding 20. However, for a sample size calculation, you might choose a more conservative (smaller) HR to ensure you have enough power. 5 o Question: What is the minimum HR you would consider meaningful to detect? For this exercise, shall we assume a conservative but still strong HR of 3.0? 4. Proportion of Population Exposed (Smokers): We need to estimate the prevalence of current smoking in your target population (adults aged 40-75). 10 According to sources like the CDC, the prevalence of smoking in US adults is around 12-14%. o Question: Would an estimated prevalence of 15% (p_smoker = 0.15) be a reasonable assumption for your study population? 5. Event Rate in the Unexposed Group (Never-Smokers): We need the 15 incidence rate of lung cancer in never-smokers aged 40-75. This can be found in cancer registry data (like the SEER program) or large cohort studies. A plausible estimate might be around 20 cases per 100,000 person-years. o Question: If we plan a 10-year follow-up, the probability of an event in the never-smoker group would be approximately 20 20 / 100,000 * 10 = 0.002. Does this seem like a reasonable starting point? Putting It All Together: An Example Calculation 1. Calculate required number of events (d): Calculate total sample size (N): 25 The overall probability of an event in the cohort is a weighted average of the probabilities in each group. o P(event) = (P(event|never_smoker) * p_never_smoker) + (P(event|smoker) * p_smoker) 2. Adjust for Loss to Follow-up: Over a 10-year period, some 30 participants will inevitably drop out. It is wise to inflate the sample size to account for this. If we assume a 20% loss to follow-up, the final sample size would be: HK 30136966 A 13 o Final N = 19,615 / (1 - 0.20) = 19,615 / 0.80 ≈ 24,519 Therefore, you would need to recruit approximately 24,520 participants (roughly 3,680 smokers and 20,840 never-smokers) and follow them for 10 years to achieve 80% power to detect a hazard ratio of 3.0 or greater. Please review these assumptions. The final number is highly sensitive to the HR and 5 the event rate. If we assumed a higher HR (which is realistic for smoking and lung cancer), the required sample size would be smaller. Preferably, the third output comprises one or more of guidance on the procedures for testing the difference in baseline variables, statistical assumptions required, steps to test statistical assumptions and sample size calculation. 10 Optionally, the method may further comprise verifying by the AI model executing on the processor of the server 20 the third output against standard checklists corresponding to said type of research study, or in the absence of a standard checklist identifying further information suggested. Preferably, the standard checklists are selected according to the type of research study from a group comprising checklists for observational studies, systematic review and 15 meta-analysis, interventional studies and diagnostic studies. Optionally, this may include consulting reporting checklists of the corresponding study type by tabulating the items against the generated proposal and suggesting what further information is needed to fulfill the respective criteria. For example, corresponding STROBE criteria may be used for observational studies, PRISMA for systematic review and meta-analysis, 20 CONSORT for interventional studies and STARD for diagnostic study. For other types not included above, another relevant existing framework may be suggested. If there is no corresponding reporting framework or standard, this is mentioned explicitly and the most reasonable items to cover suggested. Continuing the example: 25 To ensure transparent and complete reporting, the final manuscript will adhere to the STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) statement. The table below shows how our proposal aligns with key STROBE items. STROBE Item # Recommendation How Proposal Addresses It 1 Title / Abstract: Identify study's design. Yes, "Prospective Cohort Study". 2 Introduction: Background / rationale. Yes, section 1 covers this. HK 30136966 A STROBE Item # Recommendation How Proposal Addresses It 3 Introduction: Objectives. Yes, implicitly stated in section 1. 4 Methods: Study design. Yes, specified in section 2. 5 Methods: Setting. Yes, multicenter primary care networks. 6 Methods: Participants (eligibility, recruitment). Yes, detailed in section 2. 7 Methods: Variables (outcomes, exposures, confounders). Yes, detailed in section 3. 8 Methods: Data sources / measurement. Yes, questionnaires and registry linkage. 9 Methods: Bias. Addressed by prospective design (reduces recall bias). Selection bias to be monitored. 10 Methods: Study size. Yes, detailed in section 4. 12 Methods: Statistical methods. Yes, detailed in section 4. 14 Results: Participants (flow diagram). To be completed: Will need to track participants from recruitment to final analysis. 15 Results: Descriptive data. To be completed: Will be generated after data collection. 16 Results: Outcome data (events, person-time). To be completed: Will be generated after data collection. 17 Results: Main results (unadjusted and adjusted HRs). To be completed: Will be generated after analysis. 5 Further information needed to fulfill all criteria will be generated during the study's execution (e.g., participant flow diagram, descriptive results, main findings). A full research study may optionally be generated from the answers and intermediate outputs generated to date. This may be configured according to a further aspect of the disclosure; which may include the following approach. 1. Start from introduction of the research problem, current knowledge gap, the value of current study. 2. Methods. - State the study design, single or multicenter - Patient characteristics, inclusion and exclusion 14 10 HK 30136966 A 15 3. Data collection - Type of data to be collected - Proposed way to collect data 4. Statistical design and sample size consideration 5. Ethical / privacy potential issues and mitigation 5 6. Proposed study timeline and resource needed Optionally, as depicted further with reference to Figure 1C, if the research idea is determined to be for epidemiological study, observational study or study regarding risk factor and exposures, study involving intervention, the system may be configured to ask about the outcome to be examined and the most suitable data type to measure that outcome proposed. 10 The user may be asked to provide information about the exposure / risk factor they plan to look at, relating to the outcome of interest and the data type for the predictors. Alternatively, the most appropriate data type may be suggested. That is, if the research premise is formulated as an epidemiological study; observational study or study involving intervention, an intermediate output is generated for confirmation by the 15 user of the outcome under evaluation, exposure or risk factor to be evaluated. if the research premise is formulated as an epidemiological study type; observational study type or interventional study type, the second output further comprises a suggested suitable data type for confirmation. Alternatively, if the research question is formulated as a diagnostic test an intermediate output 20 S3A is generated by the artificial intelligence model to obtain intermediate user input S3B on the outcome of interest, user input on a standard benchmark test for measuring the outcome of the interest, and user input parameters required for execution of said diagnostic test. Preferably, in each of the method steps described above, the method of developing a research protocol each of outputs are only generated once the preceding step has been completed. 25 Preferably, outputs for each step are generated from stored earlier decisions and assumptions using an artificial intelligence model which is a Large Language Model (LLM). Advantageously, the Large Language Model (LLM) is configured by an external control architecture which requires that each of outputs are only generated once the preceding step has been completed and stores state information for each step. 30 Advantageously, the system and method of the present disclosure operates an artificial model (preferably, a large language model, LLM) to interact with a user to develop a research protocol by obtaining necessary information from the user in partnership. HK 30136966 A 16 The system and method of present disclosure assists in helping researchers gain confidence and understanding of various aspects of conceptualising a research protocol, demystifying statistical concepts, recommending statistically sound approaches, appropriate variables, appropriate tests and clarifying assumptions. The system and method of the present disclosure interacts with the user and generates output in an ongoing dialogue, and may be 5 accessible to the user over a network for example as a chatbot. As described herein the system and method of the present disclosure is a system which includes a framework which controls the behaviour of the AI model when performing biostatistical or research design tasks. The control system operates independently of the underlying generative language model and interfaces with the model through standard 10 input / output mechanisms. This means that the same control architecture can be applied to different LLM or SLM platforms; and the desired specialised behavior is achieved without retraining or modifying model parameters. The behavior of the system is based on modifications to system-level orchestration rather than changes to the base model. In contrast to giving the model long, complex instructions in a long detailed user 15 prompt / system prompt, the system and method of the present disclosure uses an operational framework which enforces a step by step workflow, where each step by various modules of the AI model must be completed before the next begins. As such, progression between stages is restricted until predefined prerequisite conditions are satisfied; and later analytical outputs are blocked if required prior information is missing or invalid. This prevents subsequent 20 modules of the model from generating downstream conclusions prematurely. As disclosed herein the system maintains a persistent, machine-readable analytical state that records completed workflow stages, stores validated decisions and assumptions, tracks unresolved dependencies, and conditions subsequent model outputs based on stored state information. This persistent state enables continuity and consistency across multi-turn 25 interactions and prevents contradictory or redundant outputs The system achieves improved output consistency across interactions, reduced logical contradictions, prevention of premature or invalid conclusions, increased methodological completeness, and improved efficiency by avoiding over-constrained prompting. These effects arise from the interaction between the prerequisite-enforced workflow, 30 minimal-instruction constraints, persistent state management, and external control architecture, rather than from the language model alone. The system and method of the present disclosure has been demonstrated to pproduce more accurate, consistent, and stable outputs than traditional prompt heavy approaches. HK 30136966 A 5 10 15 20 25 30 The above embodiments are described by way of example only. Many variations are possible without departing from the scope of the disclosure as defined in the appended claims. For clarity of explanation, in some instances the present technology may be presented as including individual functional blocks including functional blocks comprising devices, device components, steps or routines in a method embodied in software, or combinations of hardware and software. Methods according to the above-described examples can be implemented using computer- executable instructions that are stored or otherwise available from computer readable media. Such instructions can comprise, for example, instructions and data which cause or otherwise configure a general purpose computer, special purpose computer, or special purpose processing device to perform a certain function or group of functions. Portions of computer resources used can be accessible over a network. The computer executable instructions may be, for example, binaries, intermediate format instructions such as assembly language, firmware, or source code. Examples of computer-readable media that may be used to store instructions, information used, and / or information created during methods according to described examples include magnetic or optical disks, flash memory, Universal Serial Bus (USB) devices provided with non-volatile memory, networked storage devices, and so on. Devices implementing methods according to these disclosures can comprise hardware, firmware and / or software, and can take any of a variety of form factors. Typical examples of such form factors include laptops, smart phones, small form factor personal computers, personal digital assistants, and so on. Functionality described herein also can be embodied in peripherals or add-in cards. Such functionality can also be implemented on a circuit board among different chips or different processes executing in a single device, by way of further example. The instructions, media for conveying such instructions, computing resources for executing them, and other structures for supporting such computing resources are means for providing the functions described in these disclosures. Although a variety of examples and other information was used to explain aspects within the scope of the appended claims, no limitation of the claims should be implied based on particular features or arrangements in such examples, as one of ordinary skill would be able to use these examples to derive a wide variety of implementations. Further and although some subject matter may have been described in language specific to examples of structural features and / or method steps, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to these described features or acts. For example, such functionality can be distributed differently or performed in components other than those 17 35 HK 30136966 A 18 identified herein. Rather, the described features and steps are disclosed as examples of components of systems and methods within the scope of the appended claims. HK 30136966 A 1 CLAIMS 1. A method of developing a research protocol by a processor of a computing platform executing an artificial intelligence model; the method comprising the sequentially performed steps of: receiving by the processor a first user input from a user operable device comprising a research 5 idea; generating a first output by the processor from the research idea; wherein the first output comprises a proposed question for a research study; receiving confirmation of the first output from the user; generating a second output by the processor from the first output wherein the second output 10 comprises a research study framework and outcome addressing the proposed question, said research study framework having a predetermined type; receiving by the processor confirmation of the second output from the user; generating by the processor a third output from the research study framework and outcome, wherein said third output comprises a recommendation of the statistical parameters for the 15 research protocol. 2. The method of developing a research protocol according to claim 1 further comprising verifying by the model executing on the processor the third output against standard checklists corresponding to said type of research study, or in the absence of a standard checklist identifying further information suggested. 20 3. The method of developing a research protocol according to claim 2 wherein the standard checklists are selected according to the type of research study from a group comprising checklists for observational studies, systematic review and meta-analysis, interventional studies and diagnostic studies. 4. The method of developing a research protocol according to any one of the preceding 25 claims wherein the first output is generated by the processor using a machine learning model configured such that if the research idea is determined to be about comparing intervention or exposures the research question is formulated using a Population Intervention Comparison and Outcome (PICO) framework. 5. The method of developing a research protocol according to any one of the preceding 30 claims wherein if the research question is formulated as a diagnostic test an intermediate output is generated by the artificial intelligence model to obtain user input on the outcome of HK 30136966 A 2 interest, user input on a standard benchmark test for measuring the outcome of the interest, and user input parameters required for execution of said diagnostic test. 6. The method of developing a research protocol according to any one of the claims 1- 3 wherein if the research premise is formulated as an epidemiological study; observational study or study involving intervention, an intermediate output is generated for confirmation by the 5 user of the outcome under evaluation, exposure or risk factor to be evaluated. 7. The method of developing a research protocol according to claim 6 wherein if the research premise is formulated as an epidemiological study type; observational study type or interventional study type, the second output further comprises a suggested suitable data type for confirmation. 10 8. The method of developing a research protocol according to any one of the preceding claims wherein the third output comprises one or more of guidance on the procedures for testing the difference in baseline variables, statistical assumptions required, steps to test statistical assumptions and sample size calculation. 9. The method of developing a research protocol according to any one of the preceding 15 claims wherein each of outputs are only generated once the preceding step has been completed. 10. The method of developing a research protocol according to claim 9 wherein outputs for each step are generated from stored earlier decisions and assumptions using an artificial intelligence model which is a Large Language Model (LLM). 20 11. The method of developing a research protocol according to any one of the preceding claims using an artificial intelligence model which is a Large Language Model (LLM) which is configured by an external control architecture which requires that each of outputs are only generated once the preceding step has been completed and stores state information for each step. 25 12. The method of developing a research protocol according to any one of the preceding claims further comprising generating a research protocol in accordance with a predetermined framework. 13. A system for developing a research protocol, the system comprising a processor of a computing platform executing an artificial intelligence model comprising a plurality of modules: 30 comprising a research question generation module configured for generating a first output comprising a research question from a research idea received from the user; HK 30136966 A 3 a research study framework generation module for generating a second output comprising a research study framework and outcome from the first output, said research study framework having a predetermined type; a statistical parameter recommendation module for generating a third output from the second output; wherein said third output comprises a recommendation of the statistical parameters 5 for the research protocol. 14. The system for developing a research protocol comprising a processor of a computing platform executing an artificial intelligence model according to claim 13 further comprising: a verification module for verifying the third output against standard checklists corresponding to said type of research study, or in the absence of a standard checklist identifying further 10 information suggested. 15. The system for developing a research protocol comprising a processor of a computing platform executing an artificial intelligence model according to claim 13 or claim 14 further comprising: an external control architecture which ensures that outputs of each module are generated 15 sequentially only upon execution by the preceding module and user confirmation thereof, and wherein said framework is configured for ensuring state information for each module is stored. 16. The system for developing a research protocol comprising a processor of a computing platform executing an artificial intelligence model according to claim 13 or claim 14 wherein each of the modules are executed on an artificial intelligence model which is a Large Language 20 Model (LLM) which is configured by an external control architecture which requires that each of modules operate sequentially once a preceding module has generated an output and wherein state information for each module is stored. HK 30136966 A 1 Fig 1A Fig 1B HK 30136966 A 2 Fig 1C HK 30136966 A
