Because of the explosion of information, authors have a grave responsibility to ensure their work reflects good science and is reported in a clear and responsible way.

Good science demands good questions or hypotheses. What constitutes a good question or hypothesis? I suggest three criteria: (1) The question must be answerable or the hypothesis addressable, (2) The answer will change how we think, and (3) The answer will change how we behave. 

For a question to be answerable or a hypothesis to be addressable, they must be posed in terms of study variable. For prospective studies these will almost always require independent variables (the groups or cohorts) and dependent variables (the outcome measures or surrogates). For retrospective clinical studies in which there is only one group, the question or hypothesis must contain the dependent variables. Purposes or aims or objectives are almost always descriptive and typically vague in the sense they are rarely posed in terms of study variables; I suggest avoiding purposes. Posing a question or hypothesis in terms of study variables is a necessary, but not sufficient criterion to ensure an addressable question or hypothesis: The materials (the human or animal subjects or specimens or models) must be sufficient in scope to ensure generalizability ("power" when using subjects or specimens) and the methods of measurement and analysis must be reproducible and accurate (valid). Further, a question must be unambiguously answerable "yes" or "no" by the materials and methods used, or a hypothesis must be unambiguously supportable or refutable. Questions that cannot be so answered are not "good" questions and investigators should expend their efforts elsewhere.

When knowledge alone was sufficient than none of the medical doctors would have smoked. - ?inasi Özsoylu

For a question or hypothesis to change the way we think, the answer must be uncertain. Most publications in medicine merely confirm information about which we are fairly certain. For example, reporting the results of one implant that is substantially similar to another will likely yield similar results and we learn little from the report. Popper argued the best hypotheses are those about which we are least certain. From these "high informative content hypotheses" we learn the most. On the other hand, from "low informative content hypotheses" the answer to which we are fairly certain ("We hypothesize a Charnley total hip arthroplasty produces good results" or "We hypothesized that with navigation, TKA alignment would be less variable") we learn little. (I further note the first hypothesis is not posed in terms of study variables because "results" is not a variable.) 

For a question of hypothesis to change the way we behave, the answer must again be uncertain. If we are basing the way we treat patients on what we believe is the best available evidence for a given treatment and then conduct a study of that treatment, the chances of changing our treatment (the way we behave) is small. On the other hand, when the answer is uncertain, we are just as likely to find an answer contrary to the rationale for the treatment, then change the way we treat patients because of that contrary answer. 

A good question will always be preceded by tightly reasoned and compelling rationale. "One may think of formulating rationale and hypotheses as Aristotelian logic (a modal syllogism) taking the form: If A, B, and C, then D, E, or F. The premises A, B, and C, reflect accepted facts (rationale) whereas D, E, or F reflect logical outcomes or predictions (questions or hypotheses). The premises best come from published data, but when data are not available published observations (typically qualitative), logical argument, or consensus of opinion can be used. The strength of these premises is roughly in descending order from data to observations or argument to opinion. D, E, or F reflect logical consequences. For any set of observations, any number of explanations (D, E, or F) logically follow. Therefore, when formulating hypotheses (explanations), researchers designing experiments and reporting results should not be wed to a single explanation."¹  

I conceptualize hypotheses as one of two sorts: an explanatory hypothesis or a predictive hypothesis. In both cases one begins with a set of observations (accepted "facts"). From those observations we can either explain the entire set of phenomena by a single explanation (or more likely multiple explanations) or from an explanation ("theory," "model") we can predict a new phenomenon. An example of the former is observing a high rate of fatalities in workers painting radium on watch dials; the two observations are high fatality rates of unexplained deaths and the fact that they worked in a factory painting radium. An example of the latter is Einstein's theory of special relativity (an explanation based upon known observations and logic) which predicted gravity would bend light.

Most of us reporting clinical studies will not be posing explanatory hypotheses, and it would be artificial to attempt to do so. Rather, we can ask compelling questions: does treatment "A" produce greater survival than treatment "B"? There are three reasons to conduct then report a clinical study: (1) Report entirely new information; (2) Confirm previously reported but preliminary information; (3) Address or introduce a controversy in the literature. Few of us report entirely novel information; most often the literature contains closely related work (see Dr. Bernstein's contribution on originality). While it may be strictly correct that all studies are unique (more or less slight variations in materials and methods or analysis) closely related studies cannot be considered entirely novel. Rather, we far more commonly report studies which confirm related past work. An example might be comparing the healing rates of a locked nail that has been modified compared to a previous nail. Less commonly we conduct and report studies that address a controversy in the literature. Many studies arrive at contradictory conclusions, and a new study using relatively comparable methods to those studies can provide additional insight. Or, in our clinical practices we might make observations seemingly contradictory to what is generally accepted in the literature and conduct a study to introduce an alternative point of view (or controversy).

An average PhD thesis is transfering the bones from the old to the new grave. - J. Frank Dobise

Whether we formally introduce questions or hypotheses, we have a responsibility to introduce good ones. Either must be carefully reasoned and crafted. Either must meet the three criteria of a good question. 
Examples of good questions or hypotheses (meeting the three criteria): 

• Does photodynamic therapy inhibit physeal growth? (The answer is uncertain but could introduce a new way to treat angular deformities in children.) 
• Does adjunctive cryosurgery reduce the rate of recurrence of aneurysmal bone cysts? (The answer is uncertain but could change the way we treat patients.) 
• We hypothesized the rate of operative treatment and the surgeons' work intensity would not differ between payer types after accounting for diagnosis. (The answer is uncertain and could alter reimbursements.) 
• We hypothesized the failure strength of a transosseous anchor double-knot repair would exceed the mean ultimate strength of transosseous fixation. (The answer is uncertain and it could change the way we repair tendons.) 
  Examples of poor questions or hypotheses: 
• We asked whether photodynamic therapy could be used to initiate premature vascularization and calcification of growth plates. (This is not tightly reasoned because it is not clearly established that photodynamic therapy inhibits physeal growth and therefore asking mechanistic questions is premature.) 
• Do the clinical characteristics of fibrosarcoma patients with metastasis after 2 years differ from those of patients with earlier metastatic event? (What clinical characteristics - the author had not previously stated - why would those particular characteristics be important, and how would the information change how we treat patients?) 
• We hypothesized osteopontin and bone sialoprotein are involved in the inductive process. (It is well known they are and therefore the answer is certain.) 
• We hypothesized the motion between autopsy retrieved polyethylene inserts and the tibial baseplates increases with time in vivo. (The hypothesis is almost certainly true based upon what we know of fundamental behavior of composite structures and implant retrievals.)

  The main purpose of science is to disprove your hypothesis. -Karl Popper

Tips

• Ensure each question or hypothesis is preceded by at least two observations logically leading to the question (i.e., Aristotelian logic).
• Ensure all questions or hypotheses are posed in terms of explicit study variables.
• Ensure all questions are unambiguously answerable "yes" or "no".
• Ensure all hypotheses are unambiguously supportable or refutable.

Avoid

• Aims, purposes, objectives.
• Conducting studies for which the answers are fairly certain.
• Repeating studies for which there are multiple closely related studies in the literature unless those studies arrive at contradictory conclusion.