hospitals mortality rate
Locked
 
|
Q: Dear Experts,
I would like to monitor hospitals mortality rate in our hospital by using p-control chart to see if the rates are within our expectation, according to my knowledge mortality rate equation as follows, Mortality rate = no. of patients who died / no. of patients admitted. In a hospital we have a patient who admitted past month e.g. Aug and died in Aug or sometimes in Sep. My question is, if I would like to calculate mortality, rate lets say for Aug, are the data for the numerator and denominator should be for Aug only? Hope I was clear enough and many thanks in advance. Omar. --------------------------------- How low will we go? Check out Yahoo! Messengers low PC-to-Phone call rates. |
|
Postscript:
>I'd say this one is fairly clear: the August denominator (admissions) >should be the patients admitted in August. The August numerator (died) >should be the number of patients admitted in August who died, >regardless of when they died. Meaning, then, that you can't calculate the August rate until all patients admitted in August have died or been discharged. I don't know what your maximum length of stay is. I think patients staying over about 4 weeks need special handling, but my head is fuzzy tonight, and I can't recommend what. |
|
|
Alternatively, if one can't wait for all cases to resolve themselves:
Mortality rate for August = (all patients admitted in August who died)/(all patients admitted in August who died + all patients admitted in August who have been discharged) ignoring those who were admitted in August but have neither died nor been discharged as of the time of calculation. David Wasserman ----- Original Message ----- From: "Richard Ristow" <[hidden email]> To: <[hidden email]> Sent: Sunday, September 24, 2006 7:47 PM Subject: Re: hospitals mortality rate > Postscript: > >>I'd say this one is fairly clear: the August denominator (admissions) >>should be the patients admitted in August. The August numerator (died) >>should be the number of patients admitted in August who died, >>regardless of when they died. > > Meaning, then, that you can't calculate the August rate until all > patients admitted in August have died or been discharged. I don't know > what your maximum length of stay is. I think patients staying over > about 4 weeks need special handling, but my head is fuzzy tonight, and > I can't recommend what. > |
|
|
Dear experts,
Many thanks, the ideas were very valuables; I think the equation of Mr. Ristow is very difficult to implement, because in this case may be we needs more than 3 months to wait until all Augusts cases resolve (died or discharges). Omar. David Wasserman <[hidden email]> wrote: Alternatively, if one can't wait for all cases to resolve themselves: Mortality rate for August = (all patients admitted in August who died)/(all patients admitted in August who died + all patients admitted in August who have been discharged) ignoring those who were admitted in August but have neither died nor been discharged as of the time of calculation. David Wasserman ----- Original Message ----- From: "Richard Ristow" To: Sent: Sunday, September 24, 2006 7:47 PM Subject: Re: hospitals mortality rate > Postscript: > >>I'd say this one is fairly clear: the August denominator (admissions) >>should be the patients admitted in August. The August numerator (died) >>should be the number of patients admitted in August who died, >>regardless of when they died. > > Meaning, then, that you can't calculate the August rate until all > patients admitted in August have died or been discharged. I don't know > what your maximum length of stay is. I think patients staying over > about 4 weeks need special handling, but my head is fuzzy tonight, and > I can't recommend what. > --------------------------------- Do you Yahoo!? Everyone is raving about the all-new Yahoo! Mail. |
|
|
In reply to this post by Omar Farook
Are you going to request an attributes P or NP chart for this using month as your group? If so, I think that deaths are your defect count and hospital population is your sample size variable. However, since your samples are not discrete--that is a patient might be in the hospital for several months and thus participate in multiple samples--I'd worry about the limits computed. I would have thought that SURVIVAL or Kaplan-Meir is a better approach.
-----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Omar Farook Sent: Sunday, September 24, 2006 4:05 AM To: [hidden email] Subject: hospitals mortality rate Q: Dear Experts, I would like to monitor hospital's mortality rate in our hospital by using p-control chart to see if the rates are within our expectation, according to my knowledge mortality rate equation as follows, Mortality rate = no. of patients who died / no. of patients admitted. In a hospital we have a patient who admitted past month e.g. Aug and died in Aug or sometimes in Sep. My question is, if I would like to calculate mortality, rate let's say for Aug, are the data for the numerator and denominator should be for Aug only? Hope I was clear enough and many thanks in advance. Omar. --------------------------------- How low will we go? Check out Yahoo! Messenger's low PC-to-Phone call rates. |
|
|
In reply to this post by Omar Farook
At 01:55 AM 9/25/2006, Omar Farook wrote:
>I think the equation of Mr. Ristow is very difficult to implement, >because >in this case may be we needs more than 3 months to wait until all >August's cases resolve (died or discharges). Well, then, you have to decide what you MEAN by "August mortality", starting by ignoring what you can and can't measure. In one way, apart from difficulties in measuring (the "three months to wait"), the definition I gave you is correct. What's probably meaningful to the patients and their families is, if someone is admitted to this hospital, what's the probability they will (ever) walk out alive? It may not be the right measure of hospital performance, though. >Now, David Wasserman <[hidden email]> suggsted: > >>Mortality rate for August = (all patients admitted in August who >>died)/(all >>patients admitted in August who died + all patients admitted in >>August who >>have been discharged) ignoring those who have neither died nor been >>discharged. That's a good idea, but it tacitly assumes that mortality for long-stay patients is similar to that for short-stay patients; at best, that can't be relied on. Much more likely, it isn't true. (I don't think this is a survival-analysis problem, by the way. A discharged patient isn't 'censored', but favorable-outcome.) Even if you can wait for all cases to be resolved, that's a problem: long-stay and short-stay patients likely have different mortality patterns. You should look, if you haven't. What is the distribution of your lengths of stay? What's the median, the 90th percentile, the 99th percentile? For a lot of hospitals, most stays are short, with a 'tail' of stays that may be very long. If your 90th percentile is two weeks, a three-outcome model might make sense: Patients are . Discharged within two weeks . Died within two weeks . Retained for stay longer than two weeks. Changes in that last are meaningful, too. Back to the denominator, the 'measure of exposure'. THAT depends on your mortality model. One denominator for August would be the number of patient-days in the hospital in August. That would be appropriate if, say, most mortality was due to meteors falling through the roof and hitting patients. (OK, that's facetious. But it is based on the 'day' being the unit of risk - that the mortality is strongly influenced by the length of stay, itself. Try it, if you like: graph the percent of patients who are there for day 1, 2, ... of their stays, die on that day. If it's fairly constant, you have 'meteors'. But I doubt it will be. Likely, daily risk of mortality declines a lot with length of stay: many admissions are for acute conditions that resolve into discharge or death quite soon; the longer-term chronic conditions are likely to drag on, whatever the outcome.) See what I mean? Apart from the problem of waiting for outcomes, what denominator is appropriate depends on your risk model. One more problem: "case-mix adjustment." That is, mortality partly depends on how sick the patients are when they come in. That's commonly resolved by classifying patients into categories like DRGs ("diagnostically related groups"), all of whose members should be comparably 'sick'. . If your case mix varies, mortality change may reflect that, rather than anything the hospital does . Even if the case mix doesn't vary, hospitals may be 'good' on some types of case, poor on others. (As an extreme: Don't compare neonatal intensive care patients with adult heart disease patients.) You may need to break THAT down - requiring, of course, more data for reliable conclusions. There - that muddies the waters, at least. There's probably a point of view from which your question has a simple, defensible answer; but as posed, it raises more questions like these. -Good luck, Richard Ristow |
|
|
In reply to this post by Omar Farook
Omar --
As someone who does this for a living, let me give you some ideas (the comments from Richard Rostow notwithstanding). 1. Depending on the size and complexity of your hospital, I'm not sure you're going to derive much of value from a monthly analysis of morality rates. It's usually better to trend over more time, and to isolate for specific mortality causes. If you run a P-chart on a 500-bed hospital that averages 40 deaths per month, but that hospital received two dozen victims of a building fire who died in Emergency on one day, your "rate" is going to appear to be out of control for that month. Control charts help to identify the presence of special-cause variation -- so trending something without being sensitive to the variety of possible causes is going to diminish the effectiveness of a control chart. Might be better to track, e.g., mortality ascribed to delays in medication administration -- or at the least, to identify a few major groupings. 2. In general, hospital mortality rates are relatively stable over the long term, which is why a targeted QI effort based on specific causes of in-house death usually makes more sense. 3. My hospital (1,061 licensed beds over two facilities) reports monthly deaths but does not get too worked up over variation in the total. We have dedicated QI efforts based on the causes of inpatient death and track those instead (or proxy measures that can indicate situations that can lead to avoidable death, such as the surgical infection rate). This does, however, start to move far afield from SPSS, so if you have follow-up questions, please direct them to me by private e-mail. Thanks, and good luck. --- Jason E. Gillikin, CPHQ, Measurement & Evaluation Specialist Access Management Department, MC 157 [640D Towers] Spectrum Health - Grand Rapids Hospitals, 100 Michigan St. NE, Grand Rapids MI 49503-2560 Tel/616.391.1639 | Fax/616.391.3873 | Cell/269.352.5615 [hidden email] | http://www.spectrum-health.org -----Original Message----- From: SPSSX(r) Discussion [mailto:[hidden email]] Sent: Sunday, September 24, 2006 5:05 AM To: [hidden email] Subject: hospitals mortality rate Q: Dear Experts, I would like to monitor hospital's mortality rate in our hospital by using p-control chart to see if the rates are within our expectation, according to my knowledge mortality rate equation as follows, Mortality rate = no. of patients who died / no. of patients admitted. In a hospital we have a patient who admitted past month e.g. Aug and died in Aug or sometimes in Sep. My question is, if I would like to calculate mortality, rate let's say for Aug, are the data for the numerator and denominator should be for Aug only? Hope I was clear enough and many thanks in advance. Omar. --------------------------------- How low will we go? Check out Yahoo! Messenger's low PC-to-Phone call rates. |
Re: hospitals mortality rate
|
|
In reply to this post by Richard Ristow
Hi all
Just my 2 -cents (I happen to live in Europe, I think in euros, not in dollars...) Perhaps you could compute a true rate (strict definition: it involves time) instead of a simple (or complicated, as it looks) proportion: Number of cases in one month Mortality rate= -------------------------------------------------- Sum of times of every patient during that month This overcomes the problem of having to wait until a patient is dicharged, or whether he/she entered the hospital the previous month and therefore "belongs" to other month. The patient contributes to the denominator with as many days of that month he/she's been admitted. This measure is also called "density", as it gives the number of deaths per person-time unit, and can be also looked as a "speed" of death (since the denominator is time). The corresponding model for this statistic is Poisson regression. The drawback is that the rate can be interpreted only collectively, it has no meaning when applied to a single individual. HTH, Marta RR> At 01:55 AM 9/25/2006, Omar Farook wrote: >>I think the equation of Mr. Ristow is very difficult to implement, >>because >>in this case may be we needs more than 3 months to wait until all >>August's cases resolve (died or discharges). RR> Well, then, you have to decide what you MEAN by "August mortality", RR> starting by ignoring what you can and can't measure. RR> In one way, apart from difficulties in measuring (the "three months to RR> wait"), the definition I gave you is correct. What's probably RR> meaningful to the patients and their families is, if someone is RR> admitted to this hospital, what's the probability they will (ever) walk RR> out alive? RR> It may not be the right measure of hospital performance, though. >>Now, David Wasserman <[hidden email]> suggsted: >> >>>Mortality rate for August = (all patients admitted in August who >>>died)/(all >>>patients admitted in August who died + all patients admitted in >>>August who >>>have been discharged) ignoring those who have neither died nor been >>>discharged. RR> That's a good idea, but it tacitly assumes that mortality for long-stay RR> patients is similar to that for short-stay patients; at best, that RR> can't be relied on. Much more likely, it isn't true. RR> (I don't think this is a survival-analysis problem, by the way. A RR> discharged patient isn't 'censored', but favorable-outcome.) RR> Even if you can wait for all cases to be resolved, that's a problem: RR> long-stay and short-stay patients likely have different mortality RR> patterns. You should look, if you haven't. What is the distribution of RR> your lengths of stay? What's the median, the 90th percentile, the 99th RR> percentile? For a lot of hospitals, most stays are short, with a 'tail' RR> of stays that may be very long. If your 90th percentile is two weeks, a RR> three-outcome model might make sense: Patients are RR> . Discharged within two weeks RR> . Died within two weeks RR> . Retained for stay longer than two weeks. RR> Changes in that last are meaningful, too. RR> Back to the denominator, the 'measure of exposure'. THAT depends on RR> your mortality model. One denominator for August would be the number of RR> patient-days in the hospital in August. That would be appropriate if, RR> say, most mortality was due to meteors falling through the roof and RR> hitting patients. RR> (OK, that's facetious. But it is based on the 'day' being the unit of RR> risk - that the mortality is strongly influenced by the length of stay, RR> itself. Try it, if you like: graph the percent of patients who are RR> there for day 1, 2, ... of their stays, die on that day. If it's fairly RR> constant, you have 'meteors'. But I doubt it will be. Likely, daily RR> risk of mortality declines a lot with length of stay: many admissions RR> are for acute conditions that resolve into discharge or death quite RR> soon; the longer-term chronic conditions are likely to drag on, RR> whatever the outcome.) RR> See what I mean? Apart from the problem of waiting for outcomes, what RR> denominator is appropriate depends on your risk model. RR> One more problem: "case-mix adjustment." That is, mortality partly RR> depends on how sick the patients are when they come in. That's commonly RR> resolved by classifying patients into categories like DRGs RR> ("diagnostically related groups"), all of whose members should be RR> comparably 'sick'. RR> . If your case mix varies, mortality change may reflect that, rather RR> than anything the hospital does RR> . Even if the case mix doesn't vary, hospitals may be 'good' on some RR> types of case, poor on others. (As an extreme: Don't compare neonatal RR> intensive care patients with adult heart disease patients.) You may RR> need to break THAT down - requiring, of course, more data for reliable RR> conclusions. RR> There - that muddies the waters, at least. There's probably a point of RR> view from which your question has a simple, defensible answer; but as RR> posed, it raises more questions like these. |
|
|
Dear Dr. Marta,
Kindly, is the numerator of the equation represent the No of patients died in one month regardless the date of admission, the current month or the past month. Please, how could we calculate the denominator, what type of data we should look for? Any references on how to deal with this equation will be appreciated. Many thanks. Omar. Marta García-Granero <[hidden email]> wrote: Hi all Just my 2 -cents (I happen to live in Europe, I think in euros, not in dollars...) Perhaps you could compute a true rate (strict definition: it involves time) instead of a simple (or complicated, as it looks) proportion: Number of cases in one month Mortality rate= -------------------------------------------------- Sum of times of every patient during that month This overcomes the problem of having to wait until a patient is dicharged, or whether he/she entered the hospital the previous month and therefore "belongs" to other month. The patient contributes to the denominator with as many days of that month he/she's been admitted. This measure is also called "density", as it gives the number of deaths per person-time unit, and can be also looked as a "speed" of death (since the denominator is time). The corresponding model for this statistic is Poisson regression. The drawback is that the rate can be interpreted only collectively, it has no meaning when applied to a single individual. HTH, Marta RR> At 01:55 AM 9/25/2006, Omar Farook wrote: >>I think the equation of Mr. Ristow is very difficult to implement, >>because >>in this case may be we needs more than 3 months to wait until all >>August's cases resolve (died or discharges). RR> Well, then, you have to decide what you MEAN by "August mortality", RR> starting by ignoring what you can and can't measure. RR> In one way, apart from difficulties in measuring (the "three months to RR> wait"), the definition I gave you is correct. What's probably RR> meaningful to the patients and their families is, if someone is RR> admitted to this hospital, what's the probability they will (ever) walk RR> out alive? RR> It may not be the right measure of hospital performance, though. >>Now, David Wasserman suggsted: >> >>>Mortality rate for August = (all patients admitted in August who >>>died)/(all >>>patients admitted in August who died + all patients admitted in >>>August who >>>have been discharged) ignoring those who have neither died nor been >>>discharged. RR> That's a good idea, but it tacitly assumes that mortality for long-stay RR> patients is similar to that for short-stay patients; at best, that RR> can't be relied on. Much more likely, it isn't true. RR> (I don't think this is a survival-analysis problem, by the way. A RR> discharged patient isn't 'censored', but favorable-outcome.) RR> Even if you can wait for all cases to be resolved, that's a problem: RR> long-stay and short-stay patients likely have different mortality RR> patterns. You should look, if you haven't. What is the distribution of RR> your lengths of stay? What's the median, the 90th percentile, the 99th RR> percentile? For a lot of hospitals, most stays are short, with a 'tail' RR> of stays that may be very long. If your 90th percentile is two weeks, a RR> three-outcome model might make sense: Patients are RR> . Discharged within two weeks RR> . Died within two weeks RR> . Retained for stay longer than two weeks. RR> Changes in that last are meaningful, too. RR> Back to the denominator, the 'measure of exposure'. THAT depends on RR> your mortality model. One denominator for August would be the number of RR> patient-days in the hospital in August. That would be appropriate if, RR> say, most mortality was due to meteors falling through the roof and RR> hitting patients. RR> (OK, that's facetious. But it is based on the 'day' being the unit of RR> risk - that the mortality is strongly influenced by the length of stay, RR> itself. Try it, if you like: graph the percent of patients who are RR> there for day 1, 2, ... of their stays, die on that day. If it's fairly RR> constant, you have 'meteors'. But I doubt it will be. Likely, daily RR> risk of mortality declines a lot with length of stay: many admissions RR> are for acute conditions that resolve into discharge or death quite RR> soon; the longer-term chronic conditions are likely to drag on, RR> whatever the outcome.) RR> See what I mean? Apart from the problem of waiting for outcomes, what RR> denominator is appropriate depends on your risk model. RR> One more problem: "case-mix adjustment." That is, mortality partly RR> depends on how sick the patients are when they come in. That's commonly RR> resolved by classifying patients into categories like DRGs RR> ("diagnostically related groups"), all of whose members should be RR> comparably 'sick'. RR> . If your case mix varies, mortality change may reflect that, rather RR> than anything the hospital does RR> . Even if the case mix doesn't vary, hospitals may be 'good' on some RR> types of case, poor on others. (As an extreme: Don't compare neonatal RR> intensive care patients with adult heart disease patients.) You may RR> need to break THAT down - requiring, of course, more data for reliable RR> conclusions. RR> There - that muddies the waters, at least. There's probably a point of RR> view from which your question has a simple, defensible answer; but as RR> posed, it raises more questions like these. --------------------------------- Get your own web address for just $1.99/1st yr. We'll help. Yahoo! Small Business. |
Re: hospitals mortality rate
|
|
Hi Omar
Sorry for the delay, I didn't spot your message until now. OF> Kindly, is the numerator of the equation represent the No of OF> patients died in one month regardless the date of admission, the OF> current month or the past month. Yes OF> Please, how could we calculate the denominator, what type of data we should look for? It is the sum of the time (usually days) that patients stayed at hospital during that given month. Let's work with a small example: suppose you follow 5 patients (well that's a very small hospital, but it's just an example) from the first day to the last day of that month (let's say it's April, with 30 days): Patient #1: Admitted day 5, released day 20 (15 days stay, no death). Patient #2: Admitted day 3, died day 27 (24 days stay, death) Patient #3: was admitted previous month, released day 13 (follow up during that month: 13 days, no death) Patient #4: was admitted previous month, died day 28 (follow up during that month: 28 days, death) Patient #5: admitted previous month, not released at the the end of the month - still in hospital - (follow up: 30 days, no death). The death rate in April would be: Sum of deaths during that month: 2 (#2 & #4). Sum of follow up times: 15+24+13+28+30=110 person-days Death Rate = 2/110 = 0.0182 events/person-day Anyway, I expect you'll be having a message from Richard Ristow very soon (he wrote to me concerning this measure, and he didn't agree) :D This in an epidemiological approach to your problem, I'll try to find some references concerning its uses. Regards, Marta |
|
|
Dear Dr. Marta,
Good morning. Kindly, how could we explain the 0.0182 events/person-day? Many thanks. Omar. Marta García-Granero <[hidden email]> wrote: Hi Omar Sorry for the delay, I didn't spot your message until now. OF> Kindly, is the numerator of the equation represent the No of OF> patients died in one month regardless the date of admission, the OF> current month or the past month. Yes OF> Please, how could we calculate the denominator, what type of data we should look for? It is the sum of the time (usually days) that patients stayed at hospital during that given month. Let's work with a small example: suppose you follow 5 patients (well that's a very small hospital, but it's just an example) from the first day to the last day of that month (let's say it's April, with 30 days): Patient #1: Admitted day 5, released day 20 (15 days stay, no death). Patient #2: Admitted day 3, died day 27 (24 days stay, death) Patient #3: was admitted previous month, released day 13 (follow up during that month: 13 days, no death) Patient #4: was admitted previous month, died day 28 (follow up during that month: 28 days, death) Patient #5: admitted previous month, not released at the the end of the month - still in hospital - (follow up: 30 days, no death). The death rate in April would be: Sum of deaths during that month: 2 (#2 & #4). Sum of follow up times: 15+24+13+28+30=110 person-days Death Rate = 2/110 = 0.0182 events/person-day Anyway, I expect you'll be having a message from Richard Ristow very soon (he wrote to me concerning this measure, and he didn't agree) :D This in an epidemiological approach to your problem, I'll try to find some references concerning its uses. Regards, Marta --------------------------------- All-new Yahoo! Mail - Fire up a more powerful email and get things done faster. |
Re: hospitals mortality rate
|
|
Hi Omar
The key is understanding these concepts: Person-time: A measurement combining persons and time, used as a denominator in person-time incidence and mortality rates. It is the sum of individual units of time that the persons in the study population have been exposed to the condition of interest. A variant is person-distance, e.g., as in passenger-kilometers. The most frequently used person-time is person-years. With this approach, each subject contributes only as many years of observation to the population at risk as he is actually observed; if he leaves after 1 year, he contributes 1 person-year; if after 10, 10 person-years. The method can be used to measure incidence over extended and variable time periods. Rate: the number of cases developing per unit time (from: http://www.cbtrus.org/glossary/glossary2.html ) OF> Kindly, how could we explain the 0.0182 events/person-day? We can multiply & divide it by 1000: Event Rate =18.2 events /1000 person-day As I mentioned in a previous mail, this figure has no individual interpretation (we can't assimilate it to the probability a certain patient dies) but collective: we might expect 18 deaths when we have a total follow-up time of 1000 patients-days. This 1000 patients-days can be the result of 1000 patients followed 1 day, or 200 patients followed 5 days, or some patients being followed for more time than others. It has the advantage that all the information from each patient is taken into account (event if they were admitted the previous month, or hasn't be discharged by the end of this month...). It also takes into account not only the number of patients admitted, but also the time they stayed. OF> Let's work with a small example: suppose you follow 5 patients OF> (well that's a very small hospital, but it's just an example) from OF> the first day to the last day of that month (let's say it's April, OF> with 30 days): OF> Patient #1: Admitted day 5, released day 20 (15 days stay, no death). OF> Patient #2: Admitted day 3, died day 27 (24 days stay, death) OF> Patient #3: was admitted previous month, released day 13 (follow up OF> during that month: 13 days, no death) OF> Patient #4: was admitted previous month, died day 28 (follow up during OF> that month: 28 days, death) OF> Patient #5: admitted previous month, not released at the the end of OF> the month - still in hospital - (follow up: 30 days, no death). OF> The death rate in April would be: OF> Sum of deaths during that month: 2 (#2 & #4). OF> Sum of follow up times: 15+24+13+28+30=110 person-days OF> Death Rate = 2/110 = 0.0182 events/person-day HTH, Marta |
|
|
In reply to this post by Marta García-Granero
At 05:21 AM 9/30/2006, Marta García-Granero wrote:
>>OF> Kindly, is the numerator of the equation >>represent the No of patients died in one month >>regardless the date of admission, the current month or the past month. > >Yes > >>OF> Please, how could we calculate the >>denominator, what type of data we should look for? > >It is the sum of the time (usually days) that >patients stayed at hospital during that given month. > >Anyway, I expect you'll be having a message from >Richard Ristow very soon (he wrote to me >concerning this measure, and he didn't agree) :D Well, this isn't very soon - I've been off for the last couple of days. Among other things, Narragansett Bay under sail is even better than the Net. But, yes, this is a dissenting opinion. For human decision-making, measures of risk should reflect what matters to those who bear or pay for the consequences; for best understanding, they should seek sensitivity to the risk mechanisms, and insensitivity to extraneous factors. The risks of SOME ill occurrences in hospitals are broadly constant per patient-day, and increase accordingly with length of stay: hospital-acquired infections, certain kinds of medical error. Hospital performance on these should be (and is) assessed per patient-day. But hospitals exist to discharge patients alive and well, not to have them survive the most possible days in hospital. There are serious paradoxes in the patient-day measure: If average length of stay declines (as it has, dramatically, over the last decade or so), and mortality per admission declines but less rapidly, mortality per patient-day will rise despite what's overall a clear improvement. By analogy, consider air travel: Safety is not measured per passenger-mile or per passenger-hour, but per passenger. Partly, that mirrors passengers' interest: "Am I going to get off this thing alive?" Partly, it reflects the actual risks of flying, which are much more per-flight - takeoffs and landings - than per-mile or per-hour. (In all this, by the way, I've left out "case mix" issues: Patients are not all comparable, and hospital outcomes very dramatically by patient age, economic status, and particularly cause of admission. I've left out these issues, but if you're assessing hospitals, you'd better not.) |
|
|
In reply to this post by Marta García-Granero
Dear Dr. Marta,
Good morning. Suppose we have a monthly data, could we say that we might expect 18 deaths if we follow-up 33 patient 30 days? Many thanks. Omar. Marta García-Granero <[hidden email]> wrote: Hi Omar The key is understanding these concepts: Person-time: A measurement combining persons and time, used as a denominator in person-time incidence and mortality rates. It is the sum of individual units of time that the persons in the study population have been exposed to the condition of interest. A variant is person-distance, e.g., as in passenger-kilometers. The most frequently used person-time is person-years. With this approach, each subject contributes only as many years of observation to the population at risk as he is actually observed; if he leaves after 1 year, he contributes 1 person-year; if after 10, 10 person-years. The method can be used to measure incidence over extended and variable time periods. Rate: the number of cases developing per unit time (from: http://www.cbtrus.org/glossary/glossary2.html ) OF> Kindly, how could we explain the 0.0182 events/person-day? We can multiply & divide it by 1000: Event Rate =18.2 events /1000 person-day As I mentioned in a previous mail, this figure has no individual interpretation (we can't assimilate it to the probability a certain patient dies) but collective: we might expect 18 deaths when we have a total follow-up time of 1000 patients-days. This 1000 patients-days can be the result of 1000 patients followed 1 day, or 200 patients followed 5 days, or some patients being followed for more time than others. It has the advantage that all the information from each patient is taken into account (event if they were admitted the previous month, or hasn't be discharged by the end of this month...). It also takes into account not only the number of patients admitted, but also the time they stayed. OF> Let's work with a small example: suppose you follow 5 patients OF> (well that's a very small hospital, but it's just an example) from OF> the first day to the last day of that month (let's say it's April, OF> with 30 days): OF> Patient #1: Admitted day 5, released day 20 (15 days stay, no death). OF> Patient #2: Admitted day 3, died day 27 (24 days stay, death) OF> Patient #3: was admitted previous month, released day 13 (follow up OF> during that month: 13 days, no death) OF> Patient #4: was admitted previous month, died day 28 (follow up during OF> that month: 28 days, death) OF> Patient #5: admitted previous month, not released at the the end of OF> the month - still in hospital - (follow up: 30 days, no death). OF> The death rate in April would be: OF> Sum of deaths during that month: 2 (#2 & #4). OF> Sum of follow up times: 15+24+13+28+30=110 person-days OF> Death Rate = 2/110 = 0.0182 events/person-day HTH, Marta --------------------------------- All-new Yahoo! Mail - Fire up a more powerful email and get things done faster. |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |