Systematic random sampling entails random selection of only the first sample member.  The rest come after every regular interval.  Therefore it amounts to cluster sampling of one cluster.  Standard error of estimate cannot be computed.  

Consecutive sampling involves including all eligible sampling units until the sample  size is reached.  This is bound to be better than the systematic approach.  All statistics can be computed. 

Completely random would be the best choice. All available experimental units get a number, and a random number generator identifies the participants. Alternatively, randomly select 200 days in a year, and randomly select a time. Take the first experimental unit to arrive after that time. While slightly different, these two approaches should give you very equivalent results.

This is not always possible.

Say that the experimental units arrive one at a time.

1) You could flip a coin and heads you are part of the study, tails you are ignored.

2) You could roll a 6-sided die. The first roll is a 4, so you take the fourth person through the door, and then roll again. The roll is a 3, so you take the third person. And so forth.

3) You could sample on Monday. Randomly select a time interval. Say 3pm to 6pm. Within this time interval take the first 100 people and randomly select 30. Do this for the other days of the week.

4) Randomly select a day of the month, a day of the week, and a time interval. Everyone in that time interval gets sampled. Do this again next month.

Of course it is much easier to sample by going out and taking the first 273 experimental units that are close to you. However, this may give you a population that is very different than your intended target. The hospital "samples" one part of the larger human population. Some hospitals are affluent, others not so much. A hospital in McAllen Texas might get more Hispanics compared to a hospital in Jerusalem. Bus schedules might shift patient profiles or the work schedules of the people in your area. Doctor schedules would also be influential. Dr. Bob is on vacation the week you sampled, so you miss many of Dr. Bob's patients. Randomizing your sampling improves the chance that your sample applies to the population at large, rather than some minor subpopulation.

Warning: random sampling is not exactly what you want. I have 3,000 experimental units. I randomly select 200. It is possible (though not very probable) that I will get the first 200 experimental units. You might get the first 100 experimental units and the last 100 experimental units. These must be taken if you are using a random number generator and claiming that you used random sampling. Otherwise you have a "random sampling design that was biased by all the possible combinations that you didn't like." It might be more honest to maximally disperse your sample through the population (every 15th experimental unit). There is nothing random about this approach, but it avoids the shady side of claiming randomness when you don't accept a large number of the random possibilities.

You can also consider an experiment on experimental methods. Based on a large number of factors I can only sample once. I will sample the first 200 experimental units on Monday starting at 8:00. That is my experiment. It produced some interesting results. Now I go back and sample on Wednesday at 3:00 and collect another 200 experimental units. Do I get the same result? I'll do this again, and again to get 50 such experiments. It took a couple of years to do, but now I have a good idea of how that particular design interacts with the people arriving at the hospital. I can combine the data to understand the larger population. I can then advise if this is a good strategy for future work. I am also confident that my results are not an artifact of one sample of 200.

I think systematic sampling will make your research more rigor. Its recommended

In a hospital setup, sampling design must depends on the concern OPD burden and calculated sample size. I think you can consider either consecutive sampling or convenient sampling. For a probability sampling, you must have an idea about your total target population (N) which is most often not known in hospital setup that's why systematic sampling design is robust in community survey where you have the known N.

If burden is not that large in a hospital setup, you may not get the required subjects in the given tenure of two years even if you are taking all the consecutive cases. In such a situation one should perform the study in more than one center and adopt the consecutive sampling. When the burden is high, one can consider convenient sampling methods. 

Although random sampling is generally best for avoiding bias, systematic-random sampling (by classifying your data into categories, e.g., season, hospital site or doctor, etc.) may allow more statistical power and thus allow smaller N to be acceptable.

-Bob Vadas, Jr.

I appreciate the reply made by experts in the statistics and epidemiology. I would like to comments in this regards that sampling method may be decided on the basis of incidence and prevalence of disease which researcher want to recruit in hospital. Also that what are the probability of patients who can visit the concerned hospital to look for any referral bias. Referral bias itself does not allow you follow the rule for randomness in the collection of subjects for that disease therefore, equal probability of each subjects to get selected in your study may be violated. Even though, if our objective is to generalize the results only for the type of hospital where the study is being conducted as tertiary care hospital, then also researcher need to be sure that every cases which are coming to that hospital could be the sampling unit for the study. If yes, (such in case of Army Hospital) as the patients arriving to the particular hospital may not represent the overall population of the disease, therefore the systematic random sampling may not be the ideal option for the study. I think, in this case Consecutive sampling may be the ideal option for your study.

With consecutive sampling, I definitely wouldn't generalize beyond the particular season(s) when the samples are collected. There are many tradeoffs that you have to weigh in all this, and it may be useful to compare sampling methods as part of the study.

-Bob Vadas, Jr.

Nice discussion